Pear Biotech Bench to Business: Insights on Tackling Solid Tumors and Navigating Company Creation with Shelley Force Aldred

Here at Pear, we specialize in backing companies at the pre-seed and seed stages, and we work closely with our founders to bring their breakthrough ideas, technologies, and businesses from 0 to 1. Because we are passionate about the journey from bench to business, we created this series to share stories from leaders in biotech and academia and to highlight the real-world impact of emerging life sciences research and technologies. This post was written by Pear PhD Fellow Sarah Jones.

Today, we’re excited to share insights from our discussion with Dr. Shelley Force Aldred, CEO and co-founder of Rondo Therapeutics. Shelley is a serial founder and prominent figure in the antibody drug development space. 

More about Shelley:

Shelley earned a Ph.D. in genetics from Stanford where she worked on the human genome and ENCODE projects in the lab of Rick Myers. She spun her first company SwitchGear Genomics out of Stanford in 2006 with a grad school colleague who has since become her long-term business partner. After selling SwitchGear in 2013, Shelley shifted her focus from producing genomics tools to developing therapeutics: she helped build TeneoBio from the ground up, leading preclinical development of the company’s T-cell engager platform for treating liquid tumors, a platform that has generated $1.5 billion in upfront payments to date from multiple big pharmas. Shelley then moved on to start yet another company, Rondo Therapeutics, where she currently serves as CEO. There, she leads a team that develops innovative therapeutic antibodies for the treatment of solid tumors.

If you prefer listening, here’s a link to the recording!

Key takeaways:

1. The therapeutic window in immuno-oncology is narrow: tuning the immune system in the case of solid tumor treatment can be like playing with fire. To overcome this, Rondo has focused on using bispecific antibodies to find the ‘Goldilocks’ zone between efficacy and toxicity.

  • Previously at Teneobio, Shelley spearheaded efforts in preclinical development of immune cell engaging antibodies for liquid tumors. However, the lessons they learned and the molecules they developed couldn’t make a dent in solid tumors. Wanting to attack this problem head-on, Shelley and her long-time colleague Nathan Trinklein made the decision to start Rondo Therapeutics. 
  • One characteristic behavior of solid tumors that makes them particularly difficult to treat is their ability to trick the immune system into thinking they aren’t a threat. To combat this, Rondo is creating immuno-oncological therapies that can re-activate the immune cells that reside in the tumor microenvironment. 
  • Rondo’s efforts have been focused on the development of bispecific antibodies which are Y-shaped molecules with two arms that each can grab on and bind to different substrates. Rondo engineers these antibodies so that one arm recognizes and binds to proteins on the tumor cells while the other arm grabs onto immune cells. This brings the cells into close proximity so that the immune cells can recognize and kill the cancer cells. 
  • Shelley noted that other strategies such as checkpoint inhibitors and antibody drug conjugates often lack efficacy in solid tumors. In addition, CAR-T and other cell therapies have shown some promising preliminary results, but they can’t be administered in an off-the-shelf manner and are difficult to scale up.

Where we felt like we fit is as an off-the-shelf solution to driving tumor and immune cell engagement in a way that’s targeted specifically to the location of the tumor and isn’t body wide.

  • However, modulating the immune response is no easy feat. If pushed too far, the immune cells can start to attack healthy cells and tissues elsewhere in the both. Rondo’s cutting-edge bispecific antibodies ‘thread the needle’ and strike a balance between sparing healthy cells and killing tumor cells. Shelley noted that Rondo has been making steady progress in preclinical development and plans to be in the clinic in 2025. 

I think within other kinds of immune cell-engaging bispecifics, what we have a reputation for and are really good at is tuning and finding the Goldilocks zone. So, we’re going to be best in class in terms of this therapeutic window.

2. The ability to pivot and change directions is critical; one of Shelley’s strengths is her ability to follow the science and rely on the advice of her team. 

  • Being a founding member of three companies is quite an accomplishment. Shelley explained that joining a new company and growing it from the ground up is simultaneously an incredible opportunity and a ‘trial by fire.’ One of the advantages of working in a small start-up is the chance to take on roles that you might not otherwise have access to. 

[In a smaller company], you get to see more pieces than you would in a larger company. Inherently, when you’re in a group of only 10 or 20 people, there’s so much more visibility into what’s happening in other groups or in other people’s responsibility spheres. It’s really hard to get that in a larger company.

  • Each member within a smaller team has more responsibility in guiding the company and achieving critical milestones. Early founders and employees have to wear a lot of different hats to solve problems and ultimately push the company forward.
  • For example, Shelley noted that she had to spend a lot of time thinking not only about the science, but also about choosing the right targets and indications to pursue.

Part of that is staying humble and realizing you might not always be choosing the right targets on the first pass. We do high-throughput genomic space discovery, and so we always have a lot of targets in the mix; we have our lead program, but we also have backup programs. Particularly in this field, targets can go cold really quickly, depending on clinical results that are coming out from other companies.

  • Shelley also emphasized the importance of finding those who are willing to ride the roller-coaster with you. Bringing in experts and team members with different strengths can help keep the company agile. There are many reasons why a pivot might be necessary, and it is important to be willing to follow the science and the market. 
  • Even at Switchgear, the first company Shelley founded, an early pivot led to their eventual success and acquisition. In the initial view for the company, they anticipated having a small number of customers placing very large orders. However, it turned out that the market was asking that they have thousands of customers, each placing small orders through an e-commerce platform. By being willing to change their vision, the company ended up being extremely successful.

3. Your team is your most valuable resource, especially early on in company creation, and it’s important to surround yourself with a supportive community and a team you fully trust. 

  • It’s not a coincidence that Shelley found herself working alongside Nathan Trinklein at three of her companies – Switchgear, Teneobio, and Rondo. After running operations at Switchgear and overseeing its acquisition, the pair found themselves wanting to transition to therapeutics to get closer to patients and into a bigger market. 

Company building is a heavy lift, and being able to do that with someone with great capability that I trust deeply has increased my enjoyment of doing this quite a bit. But I also think it’s increased our likelihood of success at every step: we both have deep respect for each other and enough confidence that we just push on each other all the time. I mean, there’s constant pressure testing of ideas and conclusions. And I think what comes out the other side is always better than it would have been if only one of our brains was attacking it.

  • Another important relationship that needs to be established early on is between the founding team and the investor syndicate. Ideally, early-stage companies will have the opportunity to choose investors that support their long-term vision for the company – though the funding environment will likely determine exactly how much of a choice a founder has.
  • At Rondo, Shelley prioritized investors who were deeply versed in therapeutics and understood the relevant risks and timelines for milestones. Biotech tends to move at a slower pace, and finding firms that understand this can make a huge difference in the long run.

I am grateful for my current funding syndicate at Rondo… they are all really experienced therapeutics investors, and this is important because it means they have realistic expectations about what we’re going to be able to achieve on what timelines and what amount of capital this is going to take. It also means that their deep expertise and their networks help support us quite a bit.

4. It’s not a secret that starting a company is hard, but Shelley highlights a few ways she stays motivated. Explaining that she prioritizes bringing high quality talent into her companies, she says it’s a good thing to ‘feel a little bit stupid, at least once a day.’ Being challenged to do better and learn more is one of her favorite things about the job. 

  • Looking back over the summation of her experience as an operator, CEO, and co-founder, Shelley acknowledged that she grew and learned a lot about herself, “realizing with the exquisite mix of joy and pain that is starting a company, it was indeed the right place for me.” 
  • Her motivation comes from keeping herself always on the steep phase of the learning curve. Instead of focusing on what she doesn’t know or isn’t able to do, she pushes herself to learn constantly from her team and to bring in people who are experts in their roles. 
  • Not only does Shelley spend a lot of time recruiting and finding good employees, she also spends a considerable amount of time finding community within the broader biotech ecosystem to help keep her motivated.

We all fight different battles in our professional careers… and my network of other entrepreneurs has kept me afloat during a fundraising process when you’ve gotten a 50th no and you’re not sure you can get up and do it again. Talking to someone who’s been there and says, ‘I know you can get up to pitch number 51,’ is really important.

  • Shelley also explained the importance of finding people with shared experiences who can support you. While she enjoys challenging herself and pushing her limits, she also gives credit to her network and support system for keeping her grounded. For example, she regularly meets with her group of women biotech CEOs, and she’s found a sisterhood of women through her HiPower women’s group where she serves as an executive member.

These groups of women have been life-saving, sanity-saving in 100 different ways: primarily because they know exactly what it feels like to operate in shoes just like mine. So whatever battles you are fighting, finding people who have fought similar battles is really important.

5. Not all technical founders make great, long-term CEOs. However, with commitment to leadership development and a willingness to learn on the go, Shelley has shown that it’s possible to grow into the role and successfully lead a company.

  • Depending on the company structure, size, sector, and investor preference, there are a number of reasons founder-CEOs may not stay at the helm of their company long-term. However, Shelley has shown that technical founders, with the right experiences and mindset, can be effective leaders. 
  • While Shelley did admit to having a propensity for leadership growing up, she has made the conscious decision to invest in her own development: she reads countless books, has an executive coach, a therapist, and a willingness to apologize for her mistakes. Her humility and compassion also foster a positive team culture at Rondo. 

What I’ve tried to do is own those mistakes, apologize, … and do better later. I think that people are really wonderful and supportive when you say, ‘you’re in a learning process, you can have some compassion for me, as I’m learning to be a better leader. Just like I will have compassion for you.’ It’s like you’re learning to do your job better, and it’s really opened the gates to excellent feedback.

  • Before stepping into her role as CEO, Shelley gained invaluable experience in operational positions as a founding team member of Switchgear and Teneobio. When asked about her decision to take the CEO role at Rondo, Shelley explained that she had always known she wanted to get a chance at the job and put herself in a position to take that opportunity. 
  • When it comes to particular skills that have helped her to be an effective CEO, she explains that her strength isn’t in one particular area she excels at; instead, she is a jack of all trades.

I’m not off the charts in any one particular area, I think my advantage is that I’m good at a lot of things. So like I said, I’m a good scientist, I’m a good operator, I’m good at managing finances, I have a pretty natural sense of managing people, and I can kind of put all these things together. I think I also am good at synthesizing information that I get from a lot of different places, I’m willing to make hard decisions, and I have a pretty high risk tolerance.

Getting to know Shelley:

In her free time, Shelley likes to travel and is a voracious reader, with a particular affinity for mystery or detective novels or historical fiction. One thing people are surprised to learn about her is that she can drive a boat much better than she can drive a car because she grew up water skiing and boating regularly with her family. 

Some advice she would give someone looking to follow in her career footsteps would be to create your own opportunities and to not waste time being miserable in your work. She noted that even in your best jobs, not every day will be a good one. However, she said if you regularly wake up and dread going to work, it’s okay to look for something else. 

Pear portfolio company BioAge Labs announces oversubscribed $170M Series D financing

Last week, Pear portfolio company BioAge Labs announced its $170M Series D round led by Sofinnova Investments, with participation from a strong syndicate of new investors including Longitude Capital, RA Capital, OrbiMed Advisors, RTW Investments, Eli Lilly, and Amgen, among others, in addition to many existing investors. 

To mark this occasion, we wanted to share more about Pear’s partnership with BioAge Labs and its co-founders, Kristen Fortney (CEO) and Eric Morgen (COO). 

Pear’s founders, Pejman and Mar, first met Kristen in 2015 through an introduction by another company founder associated with the Stanford Genome Technology Center. At that time, Kristen was a postdoc at Stanford in Professor Stuart Kim’s lab, where she studied the genetics of extreme human longevity. At the time, Kristen had published extensively in the space of genetics and longevity, but the company was merely an idea. She was pondering the question: could we use genetic information and new machine learning techniques to develop a therapy discovery platform for longevity? Kristen’s vision at the time was just as clear as it is today. 

That year, Pear invested in BioAge’s initial seed financing, and we have gone on to successively back BioAge at every subsequent round, including the Series D. 

As it’s not common for a seed-stage focused firm like ours to invest up until the Series D round, why have we continued to support BioAge?

Significant unmet need and large market opportunity in obesity and metabolic disease

The company’s lead drug program, azelaprag, addresses obesity and metabolic diseases. A staggering 40% of American adults are considered obese, and many suffer from a host of comorbidities including diabetes, heart disease, and stroke.

One of the most exciting recent medical advances has been the remarkable success of GLP-1 receptor agonist drugs in achieving dramatic weight loss in such patients, while still being generally safe and well tolerated.

With this drug class expected to eventually exceed $150 billion in sales annually, the top two developers, Eli Lilly and Novo Nordisk, have catapulted to become the first and second largest pharma companies by market capitalization (~$740B and $550B, respectively, as of mid-Feb. 2024). 

As impressive as GLP-1 drugs are, one downside is that they can result in suboptimal body composition, in that they lead to the loss of both fat and muscle. BioAge’s preclinical studies have shown that azelaprag, which is a first-in-class oral apelin receptor agonist, can enhance body composition when combined with a GLP-1 drug. In a Phase 1b study sponsored by BioAge, azelaprag prevented muscle deterioration and promoted muscle metabolism in healthy older volunteers at bedrest.

A second limitation is that oral GLP-1 drugs have so far lagged behind the injectable versions in efficacy. Of course, most patients would strongly prefer orally dosed medications over injectables. In BioAge’s preclinical studies, azelaprag combined with a GLP-1 drug has been shown to double the weight loss achieved by the GLP-1 drug alone. Because it can be orally administered and has been well tolerated, azelaprag in combination with an oral GLP-1 drug may help to close this efficacy gap.

Human-first target discovery platform enabled by multi-omic analysis of aging human cohorts 

BioAge didn’t initially begin with a focus on a lead therapeutic asset in obesity. In fact, BioAge started as a target discovery company within the longevity space, with the ambitious goal of understanding the biology of human aging in an effort to extend human lifespan and healthspan. 

Although the longevity field has recently attracted much attention and investment, not all therapeutic strategies pursued have been equally scientifically rigorous. Many approaches rely on attempting to translate into humans tantalizing life extension or rejuvenation effects obtained in model organisms with very short lifespans like nematodes and mice.

But the biology of aging differs dramatically across species, and BioAge’s unique strategy was to partner with special biobanks that collected and stored blood from cohorts of people from middle age until death and that retained associated health records. By deploying multi-omics (primarily proteomics) and AI to interrogate the factors correlating with healthy human aging, the company generated unique insights into particular therapeutic targets of interest. 

From this platform, one of the strongest targets that emerged was the peptide that azelaprag is designed to mimic – apelin. Exercise stimulates release of apelin from skeletal muscle into the blood, and in BioAge’s cohorts, middle-aged people with more apelin signaling were living longer, with better muscle function, and better brain function. Correspondingly, in mice, azelaprag protected elderly mice from muscle atrophy & preserved function in vivo

Strong leadership team, advisors, and partners

As one might imagine, the team at BioAge has grown and matured substantially since inception in 2015. The leadership team today has world-class experience across biopharma. And in pursuing its Phase 2 study of azelaprag in combination with Eli Lilly’s GLP-1/GIP drug tirzepatide (Zepbound), BioAge will receive support from Eli Lilly’s Chorus organization, including the supply of tirzepatide and clinical trial design and execution expertise.  

It’s certainly uncommon for a postdoc straight out of the lab to lead a therapeutics company until a Phase 2 clinical study. But as Kristen relayed during a fireside chat at our Pear office, she learned a lot about what she needed to know on the job progressively over time, and she was not afraid to surround herself with experts specializing in the many functional domains required to take a drug program from a target to the clinic. 

This dedication to continual self-improvement and learning has been a hallmark of the many strong founders that we are fortunate to back at Pear. We are grateful that Kristen is helping to guide the next generation of such founders as part of our Pear Biotech Industry Advisory Council.

Pear Biotech Industry Advisory Council

For these reasons, we remain excited to support BioAge Labs. We eagerly anticipate the results of its mid-stage clinical trials of azelaprag in obesity, as well as the development of additional programs nominated from its unique human aging target discovery platform. 

Pear Biotech Bench to Business: insights on the past, present, and future of synthetic biology with Dr. Jim Collins

Here at Pear, we specialize in backing companies at the pre-seed and seed stages, and we work closely with our founders to bring their breakthrough ideas, technologies, and businesses from 0 to 1. Because we are passionate about the journey from bench to business, we created this series to share stories from leaders in biotech and academia and to highlight the real-world impact of emerging life sciences research and technologies. This post was written by Pear Partner Eddie and Pear PhD Fellow Sarah Jones.

Today, we’re excited to share insights from our discussion with Dr. Jim Collins, Termeer Professor of Medical Engineering and Science at MIT. Jim is a member of the Harvard MIT Health Sciences Technology faculty, a founder of the Wyss Institute for Biologically Inspired Engineering at Harvard, and a member of the Broad Institute of MIT. His work has been recognized with numerous awards and honors over the course of his career, such as the MacArthur “Genius” Award and the Dickson Prize in Medicine.

Hailed as one of the key pioneers of synthetic biology, Dr. Collins has not only published numerous high-profile academic papers, but also has a track record of success as a founder and as an entrepreneur, co-founding companies such as Synlogic, Senti Biosciences, Sherlock Biosciences, Cellarity, and Phare Bio. If all that wasn’t enough, he’s even thrown the first pitch at a Boston Red Sox game. We were lucky to sit down and chat with Jim about his experiences and his perspective on the future of synthetic biology. 

If you prefer listening, here’s a link to the recording! 

Key takeaways:

1. At its conception, synthetic biology was simply a ‘bottom-up’ approach to molecular biology utilized by collaborative, interdisciplinary scientists. 

  • In the late 90’s, Jim’s focus in biology began to shift: rather than continuing to explore biology at the whole organism or tissue level, he found himself more excited about molecular-scale biology. After speaking with some bioengineering faculty members at Boston University who were interested in his background in physics and engineering, Jim was quickly invited to join the department. From there, his interest in designing and engineering natural networks and biological processes flourished. 
  • At that time, however, bioengineers weren’t yet able to reverse engineer biological systems and exert precise control at the molecular scale. He asked, 

Could we take a bottom-up approach to molecular biology? Could we build circuits from the ground up as ways to both test our physical and mathematical notions and also to create biotech capabilities?

  • Though it didn’t start out as a quest to launch a new scientific field, Jim’s work contributed heavily to what would become the foundation of synthetic biology. He noted the value in bringing together scientists with diverse backgrounds to work on the same problems; for example, neuroscience had greatly benefitted from the introduction of mathematical models to describe complex neural systems. In a similar way, physicists, mathematicians, and molecular biologists began to find themselves interested in the same sorts of complex biological questions that could not be answered by any one discipline alone. 
  • Jim also acknowledged that in the early days, the tools to engineer gene networks and molecular pathways did not exist, yet his team could envision a future in which gene networks could be described and designed using elegant mathematical models and a modular set of biological tools. This goal helped to propel synthetic biology into existence.

2. The ability to program genetic circuits marked the beginning of synthetic biology and allowed efforts within the field to quickly progress. 

  • One notable 1995 publication in Science authored by Lucy Shapiro and Harley McAdams that was titled ‘Circuit simulation of genetic networks’ helped to shape Jim’s efforts in programming genetic circuits. The paper explored parallels between electrical circuits and genetic circuits and used mathematical modeling to accurately describe the bacteriophage lambda lysis-lysogeny decision circuit. In this circuit, bacteriophages that have infected bacteria cells must decide whether they are going to kill the cell or remain dormant, sparing the cell’s life.
  • Such work helped to bridge the gap between bioengineering and molecular biology at a time when many bioengineers felt largely excluded from the world of molecular biology.
  • To prove that genetic engineering was possible, the Collins lab worked to develop a genetic toggle switch in the form of a synthetic, bi-stable regulatory genetic network that could be switched ‘on’ or ‘off’ by applying heat or a particular chemical stimuli. This is significant because researchers could now add well-defined genetic networks to cells in order to precisely control their behavior or output.
  • This work by Gardner et al. was published in 2000 in the prestigious scientific journal, Nature and was titled “Construction of a genetic toggle switch in Escherichia coli.” Interestingly, in the same issue of Nature, work by Mike Elowitz’s lab at Caltech also outlined the development of a synthetic gene circuit in E. coli. Their system, dubbed the ‘Repressilator,’ was also a regulatory network in which three feedback loops could oscillate over time and change the status of the cells. Basically, it was three genes in a ring where gene A could inhibit gene B, which could inhibit gene C, which could then inhibit gene A, creating an oscillatory network. 
  • This critical body of work and scientific discovery both demonstrated that genetic engineering was possible and highlighted tools and methods that could be used to modulate molecular systems. 

3. To expand the repertoire of synthetic biology, Jim has co-founded two companies, Synlogic and Senti Biosciences, that are aimed at targeting the gut microbiome and engineering the mammalian system.

  • While initial excitement for synthetic biology applications centered on biofuel generation, the small scale bioreactors were never a match for fossil fuel companies. The paradigm in synthetic biology started shifting away from biofuel generation in the early 2000s to focus on the microbiome and its role in human disease. 
  • As local venture capitalists approached Jim and asked about what could actually be done with synthetic biology, it became clear to Jim that there were two main directions he could pursue. 

One was…an opportunity to create a picks and shovels company in synthetic biology. So, coming to create additional components or capacity to address a broad range of indications and applications, be it biofuels, industrial applications, therapeutics. The second was that you could engineer microbes to be living therapeutics, and in some cases, living diagnostics.

  • Jim partnered with Tim Lu, his former student and eventual coworker at MIT, to start Synlogic. One early direction of Synlogic was tackling a rare genetic metabolic disorder, phenylketonuria (PKU), that causes the amino acid phenylalanine to build up in the body. The idea was that they could engineer a microbe that could break down this byproduct and thereby eliminate the negative effects of the disease. This approach relied on the ability of the synthetic biologist to directly harness and control cell behavior via genetic engineering. 
  • Synlogic is also working on enzymes that produce therapeutic molecules instead of degrading toxic ones. The company now has efforts in inflammatory bowel disease and Lyme disease and has partnered with Roche to advance its pipeline. 
  • By around 2015, synthetic biology had continued to grow as an academic discipline and had moved beyond microbes to mammalian cells. Jim had since moved his lab from Boston University to MIT, and it wasn’t long before he was once again collaborating with Tim Lu, this time to apply synthetic biology in a mammalian system. This marked the start of Senti Biosciences, a company aimed at creating ‘smart medicines’ using genetic circuits.

We began to consider the possibility that we could do a mammalian version of Synlogic. Could we begin to really advance the development of human cell therapy and gene therapy using synthetic genes and gene circuits to create smart medicines? Having therapeutics that could sense their environment, sense the disease state or sense the disease target and produce therapeutics in a meaningful, decision-making way… was an exciting notion.

4. Historically, a lack of support from the venture community and insufficient infrastructure have been challenges for the diagnostics space.

  • Another company Jim helped start, Sherlock Biosciences, also leverages synthetic biology but operates in the diagnostic space. Although the diagnostic space is a notoriously challenging one, Sherlock was founded with the goal of combining approaches from synthetic biology and CRISPR technology to develop next-generation molecular diagnostics for at-home tests.
  • While many of the companies started right before the COVID-19 pandemic ultimately didn’t make it long-term, the team at Sherlock was able to quickly pivot and develop a CRISPR-based COVID-19 diagnostic that gained FDA-approval in May 2020. Notably, this test was the very first FDA-approved CRISPR product. 
  • Jim explained that the difficulties facing a company trying to operate in the diagnostics space are twofold:
    • (1) there is a lack of infrastructure for things like at-home testing, point-of-care testing, or nucleic acid tests
    • (2) there is a general lack of support for diagnostic companies in the venture community
  • Diagnostics companies are essentially valued as a multiple of revenue. In contrast, therapeutic companies can be valued based on projections 10-20 years in the future without the requirement of existing revenue. Combine this with the fact that wins tend to be much larger in the therapeutics space, diagnostic discovery and development have largely been set to the side. 
  • While COVID-19 did help to bring interest to the sector, funding and infrastructure continue to limit breakthroughs in diagnostics. 

5. Desperate for new antibiotics: a combination of synthetic biology, Machine Learning (ML) and in silico modeling has so far been fruitful.

  • With a challenging funding landscape, antibiotics have also been long-neglected by VC and industry. Despite this, Jim’s team was able to secure funding through The Audacious Project, a philanthropic effort put together by TED to support their work in antibiotic discovery. The funded project involved developing deep learning based models that could both discover and design novel antibiotics against some of the world’s nastiest pathogens. In fact, the team found success when they discovered a very powerful antibiotic called halicin. 
  • Recently published in Nature, an article by the Collins lab highlights their continued efforts in the “Discovery of a structural class of antibiotics using explainable deep learning.” 
  • Jim stressed the urgency for new antibiotic development: the pipeline has been drying up, but the demand has only increased. Acquired antibiotic resistance is also a significant problem that hasn’t yet been resolved.
  • As new, powerful antibiotics are developed, they become the last-line of defense against the worst, most deadly pathogens. However, drugs used as a last-line of defense don’t make it off the shelves very often: this means that there is less financial motivation to develop particularly potent antibiotics. To address this, Jim noted that we are going to need a new financial model to sufficiently support research in this space.

6. Past the hype cycle: the synthetic biology of tomorrow.

  • The field has experienced its fair share of ups and downs. In speaking with Jim, it’s clear that the roller coaster of high expectations and disappointing failures has not diminished his excitement about the future of synthetic biology. 
  • In 2004, the initial hype cycle was centered on biofuels and their potential to replace fossil fuels. Unrealistic expectations combined with the high cost of biofuel production led to disappointment; people began to question whether or not synthetic biology could deliver. 
  • In the second hype cycle, bold claims and an attitude that synbio could solve every problem in the world led to yet another massive let-down and shift in attitude towards the field. 

I think the markets haven’t kept pace with the public statements that are being made by some of the high priests in the field. And that’s a shame. I do think synthetic biology will emerge as one of the dominant technologies of this century. Our ability to engineer biology gives us capabilities that can address many of the big challenges that we have. But it’s still going to take a lot of time, it’s still very hard to engineer biology, and biology is not yet an engineering discipline.

  • Successes in areas where biology still outcompetes chemistry have helped to put some points back on the board for synthetic biology. Increasing utilization in therapeutic development has leveraged the efficiency of biological systems and will help to pave the way for the next way of discoveries in the field. 
  • Technologies like cell-free systems also have Jim excited about the future of synthetic biology. 

Get to know Jim Collins: 

Early career and developing a passion for science: 

  • Jim comes from a family of engineers and mathematicians and has always found himself wanting to do science. Jim explained that when he was four years old, his dad was a part of a team that designed an altimeter for Apollo 11. 
  • Another seminal event that influenced Jim’s decision to become a scientist was the decline of his grandfather’s health after a series of strokes left him hemiplegic. After watching someone he loved not receive the care or have treatment options that could restore function, Jim was inspired to pursue biomedical engineering. 
  • Once he realized that he could interface with clinicians, entrepreneurs, and policy-makers as a professor, he realized that was the path for him.

Advice for early-stage founders:

  • Find a strong business team early on to help find market fit and to guide the development of your final product. Young scientists are not trained to be good CEO’s, and it’s often challenging to navigate these decisions if you don’t have the experience.
  • Make sure your strategy has a real market pull and is differentiated from other approaches.  

Perspectives in AI with Kamil Rocki, Head of Performance Engineering at Stability AI

At Pear, we recently hosted a Perspectives in AI fireside chat with Kamil Rocki, Head of Performance Engineering at Stability AI. We discussed breakthroughs at the hardware-software interface that are powering generative AI. Kamil has extensive experience with GPU hardware and software programming from his PhD research and his work at IBM, Nvidia, Cerebras, Neuralink, and of course now StabilityAI. Read a recap of that conversation below:

Aparna: Kamil, thank you for joining us. You’ve accomplished many amazing things in your career, and we’re excited to hear your story. How did you choose your career path and what led you to work on the projects you’ve been involved with?

Kamil: My journey into the world of technology began in my 20s. After a few years of rigorous mathematical studies, I found myself in a robotics lab. I was tasked with enabling a robot to solve a Rubik’s cube. The challenge was to detect the cube’s location in an image captured by a camera, and this had to be done at a rate of 100 frames per second. 

I was intrigued by the work my peers were doing in computer graphics using Graphics Processing Units (GPUs). They were generating landscapes and waves, manipulating lighting, and everything was happening in real-time. This inspired me to use GPUs to process the images for my project.

The process was quite challenging. I had to learn OpenGL from my friends, write images to the GPU, apply a pixel shader, and then read data back from the GPU. Despite the complexity, I was able to exceed the initial goal and run the process at 200 frames per second. I even developed a primitive version of a neural network that could detect the cube’s location in the image.

In 2008, around the time I graduated, CUDA came out and there was a lot of excitement around GPUs. I wanted to continue exploring this field and heard about a supercomputer being built in Japan based on GPUs and ended up doing a PhD in supercomputing. During this time, I worked on an algorithm called Monte Carlo Tree Search, deploying it on a cluster of 256 GPUs. At that time, not many people were familiar with GPU programming, which eventually led me to the Bay Area and IBM Research in Almaden.

I spent five years at IBM Research, then moved to the startup world. I had learned how to build chips, design computer architecture, and build computers from scratch. I was able to go from understanding the physics of transistors to building a software stack on top of that, including an assembler, compiler, and programming what I had built. One of my goals at IBM was to develop a wafer scale system. This led me to Cerebras Systems, where I co-designed the hardware. Later I joined Neuralink and then Nvidia, where I worked on the Hopper architecture. I joined Stability, as we are currently in a transition to Hopper GPUs. There is a significant amount of performance work required, and with my extensive experience with this architecture, I am well-equipped to contribute to this transition.

Aparna:  GPUs have become one of the most profitable segments of the AI value chain, just looking at Nvidia’s growth and valuation. GPUs are also currently a capacity bottleneck. How did we arrive at this point? What did Nvidia, and others, do right or wrong to get us here?

Kamil: Nvidia’s journey to becoming a key player in the field of artificial intelligence is quite interesting. Initially, Nvidia was primarily known for its Graphics Processing Units (GPUs), which were used in the field of graphics. A basic primitive in graphics involves small matrix multiplication, used for rotating objects and performing various view projection transformations. People soon realized that these GPUs, efficient at matrix multiplications, could be applied to other domains where such operations were required.

In my early days at the Robotics Lab, I remember working with GPUs like the GeForce 6800 series. These were primarily designed for graphics, but I saw potential for other uses. I spent a considerable amount of time writing OpenGL code to set up the entire pipeline for simple image processing. This involved rasterization, vertex shader, pixel shader, frame buffer, and other complex processes. It was a challenging task to explore the potential of these GPUs beyond their conventional use.

Nvidia noticed that people were trying to use GPUs for general-purpose computing, not just for rendering images. In response, they developed CUDA, a parallel computing platform an application programming interface model. This platform significantly simplified the programming process. Tasks that previously required 500 lines of code could now be achieved with a program that resembled a simple C program. This opened up the world of GPU programming to a wider audience, making it more accessible and flexible.

Around 2011-12, the ImageNet moment occurred, and people realized the potential of scaling up with GPUs. Before this, CPUs were the primary choice for most computing tasks. However, the realization that GPUs could perform the same operations on different data sets significantly faster than CPUs led to a shift in preference. This was particularly impactful in the field of machine learning, where large amounts of data are processed using the same operations. GPUs proved to be highly efficient at performing these repetitive tasks.

This realization sparked a self-perpetuating cycle. As GPUs became more powerful, they were used more extensively in machine learning, leading to the development of more powerful models. Nvidia continued to innovate, introducing tensor cores that further enhanced machine learning capabilities. They were smart in making their products flexible, catering to multiple markets including graphics, machine learning, and high-performance computing (HPC). They supported FP64 computation, graphics, and tensor cores, which could be used for ray tracing and FP64. This adaptability and flexibility, combined with an accessible programming model, is what sets Nvidia apart in the field.

In the span of the last 15 years, from 2008 to the present, we have seen a multitude of different architectures emerge in the field of machine learning. Each of these architectures was designed to be flexible and adaptable, capable of being executed on a GPU. This flexibility is crucial as it allows for a wide range of operations, without being limited to any specific ones.

This approach also empowers users by not restricting them to pre-built libraries that can only run a single model. Instead, it provides them with the freedom to program as they see fit. For instance, if a user is proficient in C, they can utilize CUDA to write any machine learning model they desire.

However, some companies have lagged behind in this regard. Their mistake was in not providing users with the flexibility to do as they please. Instead, they pre-programmed their devices and assumed that certain architectures would remain relevant indefinitely. This is a flawed assumption. Machine learning architectures are continuously evolving, and this is a trend that I foresee continuing into the future.

Aparna: Could you elaborate more on the topic of special purpose chips for AI? Several companies, such as SambaNova Systems and Cerebras, have attempted to develop these. What, in your opinion, would be a successful architecture for such a chip? What would it take to build a competitive product in this field? Could you also shed some light on strategies that have not worked well, and those that could potentially succeed?

Kamil: Reflecting on my experience at Cerebras Systems, I believe one of the major missteps was the company’s focus on building specialized kernels for specific architectures. For instance, when ResNet was introduced, the team rushed to develop an architecture for it. The same happened with WaveNet and later, the Transformer model. At one point, out of 500 employees, 400 were kernel engineers, all working on specialized kernels for these architectures. The assumption was that these models were fixed and optimized, and users were simply expected to utilize our library without making any changes.

However, I believe this approach was flawed. It did not take into account the fact that architectures change frequently. Every day, new research papers are published, introducing new models and requiring changes to existing ones. Many companies, including Cerebras, failed to anticipate this. They were so focused on specific architectures that they did not consider the need for flexibility.

In contrast, I admire NVIDIA’s approach. They provide users with tools and allow them to program as they wish. This approach is more successful because it allows for adaptability. Despite the progress made by companies like Cerebras, Graphcore, and others, I believe too much time and effort is spent on developing prototypes of networks, rather than on creating tools that would allow users to do this work themselves.

Even now, I see companies building accelerators for the Transformer architecture. I would advise these companies to rethink their approach. They should aim for flexibility, ensuring that their architecture can accommodate changes. For instance, if we were to revert to recurrent nets in two years, their architecture should still be programmable.

Aparna: Thank you for your insights. Shifting gears, I’d like to talk about your work at Stability. It’s an impressive company with a thriving open-source community that consistently produces breakthroughs. We’ve observed the quality of the models and the possibilities with image generation. Many founders are creating companies using Stability’s models. So, my question is about the future of this technology. If a founder is building in this space and using your models as a foundation, where do you see this foundation heading? What’s the future of image generation technology at Stability?

Kamil: The potential of technology, particularly in the field of artificial intelligence, is immense. Currently, we’re seeing significant advancements in image generation models. The quality of these generated images is often astounding, sometimes creating visuals that are beyond reality, thereby accelerating creativity and content creation. We’re now extending this capability into 3D and video space. We’re actively working on models that can generate 3D scenes or objects and extend to video space. Imagine a scenario where you can generate a short clip of a dog running or even create an entire drama episode from a script.

We’re also developing audio models that can generate music. This can be combined with video generation to create a comprehensive multimedia experience. These applications have significant potential in the entertainment industry, from content generation for artists to the movie industry and game engine development.

However, I believe the real breakthrough will come when we move towards more industrial applications. If we can generate 3D representations and add video to that, we could potentially use this technology to simulate physical phenomena and accelerate R&D in the manufacturing space. For instance, generating an object that could be printed by a 3D printer. This could optimize and accelerate prototyping processes, potentially revolutionizing supply chains.

Recently, I was asked if a space rocket could be designed with generative AI. While it’s not currently feasible, the idea is intriguing and could potentially save a lot of money if we could solve complex problems using this technology.

In relation to hardware, I believe that generative AI and language models can be used to accelerate the discovery of new kinds of hardware and for generating code to optimize performance. With the increasing complexity and variety of models and architectures, traditional approaches to optimizing code and performance modeling are struggling. We need to develop more automated, data-driven approaches to tackle these challenges.

Aparna: You’ve broadened our understanding of the potential of generative AI. I’d like to delve deeper into the technical aspects. As the head of Performance Engineering at Stability, could you elaborate on the challenges involved in building systems that can generate video and potentially manufacture objects without error, performing exactly as intended?

Kamil: From a performance perspective, the issue of being limited by computational resources is closely related to the first question. At present, only a few companies can afford to innovate due to the high costs involved.  

This situation might actually be beneficial as it could spark creativity. The scarcity of resources, particularly GPUs, could trigger innovations on the algorithmic side. I recall a similar situation in the early days of computer science when people were predicting faster clock speeds as the solution to performance issues. It was only when they hit a physical limit that they realized the potential of parallelization, which completely changed the way people thought about performance.

Currently, the cost of building a data center for training state-of-the-art language models is approaching a billion dollars, not including the millions of dollars required for training. This is not a sustainable situation. I miss the days when I could run models and prototype things on a laptop.

One of the main problems we face is that we’ve allowed our models to become so large, assuming that compute infrastructure is infinite. These larger models are becoming slower because more time is spent on moving data around rather than on the actual computation. For instance, when I was at Nvidia, anything below 90% of the so-called ‘speed of light’ was considered bad. However, in many cases, large language models only utilize about 30-40% of the peak performance that you can achieve on a GPU. This means a lot of compute power is wasted.

People often overlook this issue. When I suggest optimizing the code on a single GPU and running it on a small model before scaling up, many prefer to simply run it on multiple GPUs to make it faster. This lack of attention to optimization is a significant concern.

Aparna: As we wrap up, I’d like to pose a final question related to your experience at Neuralink, a company focused on brain-to-robot interaction. This technology has potential applications in assisting differently-abled individuals. Could you share your perspective on this technology? When do you anticipate it will be ready, and what applications do you foresee?

Kamil: My experience at Neuralink was truly an exciting adventure. I had the opportunity to work with a diverse team of neuroscientists, physicists, and biologists, all of whom were well-versed in computing and programming. Despite the initial intimidation, I found my place in this team and contributed to some groundbreaking work.

One of the primary challenges we aimed to address at Neuralink was the communication barrier faced by individuals whose cognitive abilities were intact, but who were physically unable to express themselves. This issue is exemplified by renowned physicist Stephen Hawking, who could only communicate by typing messages very slowly using his eyes.

Our initial project involved training macaque monkeys to play a Pong game while simultaneously feeding data from their motor cortex. This allowed us to decode brain signals and enable the monkeys to control something on the screen. Although it may not seem directly related to human communication, this technology could potentially be used to control a cursor and type messages, thus bypassing physical limitations.

We managed to measure the information transfer rate from the brain to the machine in bits per second, achieving a rate comparable to that of people typing on their cell phones. This was a significant milestone and one of the first practical applications of our technology. It could potentially benefit individuals who are paralyzed due to spinal injuries, enabling them to communicate despite their physical limitations.

However, our work at Neuralink wasn’t limited to decoding brain signals and reading data. We also explored the possibility of stimulating brain tissue to induce physical movements or visual experiences. This bidirectional communication could potentially allow individuals to interact with computers more efficiently, bypassing the need for physical input devices. It could even pave the way for a future where VR goggles are obsolete, as we could stimulate the visual cortex directly. However, the safety of these techniques is still under investigation, and it’s crucial that we continue to prioritize this aspect as we push the boundaries of what’s possible.

There’s a significant spectrum of disorders that this technology could address, particularly for individuals who struggle with mobility or communication. We were also considering mental health issues such as depression, insomnia, and ADHD. One of the concepts we were exploring is the ability to read data from the brain, identify its state, and stimulate it. This could potentially serve as a substitute for medication or other forms of treatment.

However, it’s important to note that the technology, while progressing, is not entirely clear-cut. The safety aspect is crucial and cannot be ignored. At Neuralink, we’ve done a remarkable job ensuring that everything we develop is safe, especially considering these devices are implanted in someone’s head.

When we consider brain stimulation, we must also consider potential negative scenarios. For instance, if we stimulate a certain region of the brain to alleviate depression, we could inadvertently create a dependency, similar to injecting dopamine. This could potentially lead to a loop where the individual becomes addicted to the stimulation. It’s a complex issue that requires careful consideration and handling.

In addressing these challenges, we’ve engaged in extensive conversations with physicians, neuroscientists, and other experts. While some companies may have taken easier paths, potentially compromising safety, we’ve chosen a more cautious approach. Despite the slower progress, I can assure you that whatever we produce will be safe. This commitment to safety is something I find particularly impressive.

Kamil: For those interested, it’s worth noting that Neuralink is currently hiring. They’ve recently secured another round of funding and are actively seeking new talent. This is indeed a glimpse into the future of technology.

Aparna: Earlier, you mentioned an intriguing story about monkeys and reading their brainwaves. This story is related to the AI that’s been implanted in their brains and how it communicates. Could you elaborate on what happens with the models in this context?

Kamil: In our initial approach to decoding brain signals, we utilized a simple model. We had a vector of 1024 electrodes and our goal was to infer whether the monkey was attempting to move the cursor up, down, or click on something. We used static data from what we termed a pre-training session, which was essentially data recorded from the implant. The model was a two-layer perceptron, quite small, and could be trained in about 10 seconds. However, the brain’s signal distribution changes rapidly, so the model was only effective for about 10 to 15 minutes before we observed a degradation in performance. This necessitated the collection of new data and retraining of the model.

Recently, Neuralink has started exploring reinforcement learning-based approaches, which allow for on-the-fly identification and retraining of the model on the implant. During my time at Neuralink, my focus was primarily on the inference side. We trained the model outside the implant, and my role was to make the inference parts work on the implant. This was a significant achievement for us, as we were previously sending data out and back in. Given our battery limitations, performing tasks on the implant was more cost-effective. The ultimate goal was to move the entire training process to the implant.

Every day, our brains produce varying signals due to changes in our moods and environments. These factors could range from being in a noisy place, feeling tired, or engaging in different activities. This results in a constantly shifting distribution of brain signals, which presents a significant challenge. This phenomenon is not only applicable to the brain but also extends to other applications in the medical field.

Aparna: We’ve discussed a wide range of topics, from hardware design to image and video generation, and even brainwaves and implant technology. Thank you so much for these perspectives Kamil!

Thank you to Kamil for his perspectives on these exciting AI topics. To read more about Pear’s AI focus and previous Perspectives in AI talks, visit this page.

Pear Biotech Bench to Business: insights on generative AI in healthcare and biotech with Dr. James Zou

Here at Pear, we specialize in backing companies at the pre-seed and seed stages, and we work closely with our founders to bring their breakthrough ideas, technologies, and businesses from 0 to 1. Because we are passionate about the journey from bench to business, we created this series to share stories from leaders in biotech and academia and to highlight the real-world impact of emerging life sciences research and technologies. This post was written by Pear Partner Eddie and Pear PhD Fellow Sarah Jones.

Today, we’re excited to share insights from our discussion with Dr. James Zou, Assistant Professor of Biomedical Data Science at Stanford University, who utilizes Artificial Intelligence (AI) and Machine Learning (ML) to improve clinical trial design, drug discovery, and large-scale data analysis. We’re so fortunate at Pear to have James serve as a Biotech Industry Advisor for us and for our portfolio companies.

James received his Ph.D. from Harvard in 2014 and was a Simons Research Fellow at UC Berkeley. Prior to accepting a position at Stanford, James worked at Microsoft and focused on statistical machine learning and computational genomics. At Stanford, his lab focuses on making new algorithms that are reliable and fair for a diverse range of applications. As the faculty director of the Stanford AI for Health program, James works across disciplines and actively collaborates with both academic labs and biotech and pharma. 

If you prefer listening, here’s a link to the recording!

Key Takeaways:

1. James and his team employed generative AI not only to predict novel antibiotic compounds, but also to produce a facile ‘recipe’ for chemical synthesis, representing a new paradigm of drug discovery.

  • Antibiotic discovery is challenging for at least a couple of reasons: reimbursement strategies do not incentivize companies to pour resources, time and money into R&D, and antibiotic resistance has made it difficult to create lasting, efficacious products. To accelerate discovery and meet the critical need for new antibiotics, James and his team created a generative AI algorithm that could generate small molecules that were predicted to have high activity. Not only could the model generate chemical structures, but it could also produce the instructions for chemists to make these compounds. By streamlining the process from structure generation to synthesis, James and his team were able to identify a potent antibiotic for pathogens that have developed resistance to existing antibiotics. 
  • These ‘recipes’ that the algorithm generated laid out step-by-step instructions for over 70 lead compounds. After synthesizing and testing the molecules, they found that they achieved a hit rate above 80% and could synthesize 58 novel compounds. Of these 58, they found that six were validated as promising drug candidates. In addition, the model prioritized hits that had robust synthetic protocols and were predicted to have low toxicity. In this way, the model could prioritize certain small molecule features and generate both novel structures and complete recipes. 
  • The generative model in this case used a Monte Carlo Tree search to come up with recipes for new small molecules. The same logic flow and reasoning can easily be applied to other settings. For example, James and his team are working with Stanford spin-outs to apply the algorithm to other diseases such as fibrosis or for applications requiring new fluorescent molecules. 

I think it’s a good, reasonable model for AI and biotech in general to have this close feedback loop, where we start off with some experimental data. In our case, we have some experimental screening data, the actual data used to train the models, and then the model will produce some candidates or some hypotheses. Then we tried to have a more rapid turnaround to do the additional experiments … to further validate the AI’s reasoning and thinking.

2. James’ group has also used generative AI to design clinical trials that aim to be faster, cheaper, more diverse, and more representative.

  • Clinical trials are a critical bottleneck in the pipeline of therapeutic development. Patient enrollment is slow and labor intensive, and  trial design can be biased against underrepresented groups or may exclude patients left out based on criteria that don’t necessarily relate to trial outcome or patient response. To outline trial criteria, a ‘monstrosity’ of a document must be generated to explicitly lay out the rules that will be followed in patient recruitment. 
  • James noted that it’s very hard to balance all of the complicated factors required for a successful trial design. With collaborators at Genentech, James and his team have worked to develop a Generative AI algorithm called Trial Pathfinder that uses historical clinical trials and outcomes to create an optimized trial design, often including a more diverse patient population. Among groups that see more representation and inclusion in AI-generated clinical trials are women, elderly patients, patients from underrepresented groups, and patients who might be a bit sicker. These patients often end up responding just as well without experiencing the predicted adverse events. 
  • When James first started partnering with Genentech, his first goal was to understand the pain points in clinical trial design. He learned that many trials are actually very narrow and essentially recruit for the so-called “Olympic athletes” of patients. He noted one study actually did try to recruit Kobe Bryant and several Olympians. To make clinical trial outcomes more inclusive and representative of a diverse range of patients, one strategy is the use of algorithms that account for such biases. 

If we look more closely at how people design the protocols for clinical trials, often it is based on domain knowledge. But it’s also often quite heuristic and anecdotal, which is why a lot of different teams from different pharma companies–even if they’re looking at drugs of similar mechanisms–often end up with quite different trials and trial designs.

3. AI/ML in biology is only possible because biology is becoming increasingly data driven.

  • The data that we can gather from biological systems is becoming more and more rich and diverse. For example, high resolution spatial transcriptomics or single molecule experiments can now be captured alongside more traditional measurements such as RNA sequencing. Perturbation of biological systems with CRISPR/Cas-9 technology also enables a whole new suite of data that represents an area ripe for AI.
  • Even in the past year, James noted that he has seen tremendous advances in large language models and foundation models on the AI side. It’s also been interesting to see how these advances have enabled new insights in the biotech space. For example, high-throughput perturbation data collected at the single-cell resolution is something that can be extremely compatible with large language models. So far, collaborative efforts between AI and biotech have been extremely fruitful. 
  • As a professor, James and his lab are always pursuing new and exciting research directions. In particular, one project that he’s excited about is fueled by recent progress in spatial biology. As we’ve seen interest skyrocket in single-cell transcriptomics and genomics, researchers have generated huge amounts of data that have led to a variety of different findings and results. However, single cells also reside in different neighborhoods, or local microenvironments. Understanding disease and healthy states in the context of groups of cells may unlock even more insight into how patients may respond to therapies. 
  • Another promising direction James highlighted was the use of large language models to allow researchers to synthesize and analyze information across biological databases. Currently, data is often siloed, and the level of expertise required to utilize individual data sets makes it challenging to work across fields and specialties. 
  • “But this is where we think that language models can really be a unifying framework that can then help us to access data and integrate data from all these different modalities on different knowledge bases. So that’s another thing we’re working on with my students.”

This is why we’re excited about techniques and ideas like large language models that harness data from different modalities, databases, and knowledge bases to help biomedical researchers make faster innovations.

4. Landmark results at the intersection of biology and engineering, such as the Human Genome Project and the discovery of Yamanaka factors for reprogramming stem cells, motivated James to pursue a career in academia and to start his lab at Stanford.

  • James always gravitated towards math and science and started his academic path with an undergraduate major in math. Although he didn’t start out with a focus in biology, he began spending a good deal of time at the Broad Institute during his time at Harvard. At that time, many great biotechnologists were working on projects like sequencing the human genome. James began to learn more about interesting problems at the intersection of AI and biotech. 
  • As he learned about the discovery of Yamanaka factors for reprogramming induced pluripotent stem cells (iPSCs), James was fascinated by the idea that you could take these ‘biological computer programs’ and with only a few instructions, change the state of a cell. Essentially, the biological system became not just something you could study, but something you could engineer.

5. Translation of academic projects needs to happen thoughtfully in collaboration with the right partners.

  • Translation of academic projects can take a few different forms, and not every project is right for translation. James gave a few examples of projects that he knew could move beyond the Stanford campus. One project involved the development of an AI system for assessing heart disease based on cardiac ultrasound videos, or echocardiograms. Millions of these videos are collected each year in the US alone, so it’s one of the most routine and accessible ways to assess cardiovascular disease. 
  • James and his students developed an AI system to help look at these videos and determine outcomes to help clinicians make more accurate diagnoses. Not only did the work result in two publications in the prestigious scientific journal, Nature, but it progressed to clinical trials. 
  • James explained that he clearly sees the larger impact of his work; he wants not only to publish really good papers, but also to work closely with biotech and pharma companies or tech companies to ensure his findings and algorithms can actually impact human health. His focus on translation has led to at least four companies that have been spun out of his group.  
  • “At least one of the things we are doing is developing these algorithms or coming up with some of these potential drug candidates, and to really take it to the next level, either as a viable drug, a platform, or a device that we can take to patients, is something often beyond the scope of an individual PhD.”
  • James enjoys leveraging the community and resources of the Chan-Zuckerberg Biohub, a non-profit initiative aimed at bringing together interdisciplinary leaders to advance our ability to observe and analyze biological systems. Opportunities and communities such as these played a large role in drawing James to the Bay Area. He has actively sought out projects through which he can collaborate with biotech and pharma, investors, and the start-up community.

At that point, it’s really important work. [Getting FDA approval] also requires more resources. That’s where it makes sense for us to have a company that is co-founded by my students. So, for us, in this case, it’s a perfect synergy between doing the early-stage research and development, developing the algorithms to the initial validations and then having the company take over and do the submissions and then do the scaling.

Get to know Dr. James Zou: the person behind the science

James and his wife love to take advantage of all the great outdoors in the Bay Area. Every week, they go on a hike and spend a lot of time swimming or biking. One thing people may be surprised to learn about James is that he used to moonlight as a theater and restaurant reviewer. When he was living in Europe, he would write reviews of movies or restaurants for local English-language newspapers. 

For someone wanting to pursue a similar career to his, James says that it’s a very exciting time to be at the intersection of biology and AI. He encourages students in the space to develop a core technical strength in either field and then begin to explore the synergy between disciplines. 

The vertical software data gold mine 

Automation opportunities from vertical software’s data gold mine mean it has never been a better time to create purpose-built operational tools for overlooked industries. Plus, our market map of AI-enabled vertical software new entrants. 

Vertical software founders build applications across many industries but with a common mission: to empower expert operators and owners to streamline and grow their businesses with software purpose-built for their own industry. 

Over the past decade, Pear has supported founders building powerful applications for industries ranging from supply chain (Expedock, Beyond Trucks), construction (Gryps, Miter, Doxel), energy (Aurora, Pearl Street) and insurance (Federato) to home services (Conduit), travel (JetInsight, Skipper), agriculture (FarmRaise, Lasso), real estate (Hazel) and live events (Chainpass) – and more. These companies and plenty of others we admire outside of our portfolio have created hundreds of billions of dollars in enterprise value for themselves and for their customers. 

Today, with artificial intelligence capabilities easier than ever to deploy in B2B products, we think it has never been a better time to build vertical software that automates core aspects of customer operations. 

Any new vertical software opportunity — any sector that is not well-served by purpose-built software for industry-specific workflows — now looks even more promising. And, many existing vertical software tools have a rare opportunity to ramp up ACVs, increase stickiness, and build a powerful moat. 

To understand why we might be on the verge of a golden age of intelligently-automated vertical software, we’ll take a look at two problems: First, the ACV problem that has limited vertical software opportunities. Then, the defensibility problem afflicting new AI application entrants. 

The ACV Problem 

Vertical software companies often sell into fragmented industries with many small-to-medium businesses. These businesses operate on thin margins and have limited ability to pay for new operational software. Contract values – and therefore market size – for many verticals have historically been constrained, which means that many sectors remain underserved by purpose-built, cloud-based software.

Historically, vertical software companies have sought to increase the value of their customer relationships by adding bolt-on monetization features like payments or banking. Other vertical software companies have initially sold to the fragmented base of a sector and steadily added operational features to move upmarket to larger customers within their industry. 

In either case, the fundamental margin structure of the end customer remains unchanged, and few vertical software companies can reliably claim to impact their customers’ profitability in a major way. A vertical software customer might love their software’s intuitive UI, centralized system of record, navigable scheduling tools, and modern payment processing system. But these benefits rarely reduce the customer’s overall operating cost in a significant way. 

The Defensibility Problem

Despite the excitement over applications of large language models in late 2022 and early 2023, many initial products were dismissed as “thin wrappers” over an off-the-shelf model. Critics argued that these initial applications lacked product differentiation and long-term defensibility.

At Pear, we believe that proprietary data is one of the keys to a defensible AI application. Proprietary data behind B2B AI applications comes in three forms: 

Many early tools built over groundbreaking LLMs offered no form of proprietary data. At best, some products built over lightly-adapted models offered bronze or silver-level data-based defensibility. But we’ve been on the hunt for game-changing applications that build an advantage in proprietary model-training data from the start.  

The Vertical Software Data Gold Mine 

Traditional vertical software products generate enormous amounts of gold-level data, capture substantial silver-level information, and often themselves aggregate bronze-level data within their products.

Customer business logic flows through vertical software features. From operational decisions and administrative record-keeping to sales and product performance, customers of vertical software deposit reams of data daily into a rich system of record. 

The most impactful B2B applications in the next decade will rigorously structure, mine, and harness gold-level product-generated data to enable workflow automation for their end customers.

Any business process with decisions and steps encoded in a vertical software feature set will be a candidate for automation. We’re most excited about automation that enables faster information processing tied to sales growth or cost-saving opportunities for an end customer: instant diagnosis and repair commissioning for field technicians, predictive inventory capabilities embedded in B2B marketplaces, copilot-style knowledge bases that help small business owners understand the impact of every decision on their bottom line. 

Automation across many business functions will mean that vertical software companies can finally impact their customers’ margin structure – and as a result, help these customers break any linear scaling trap they face when they otherwise expand their business.

Early entrants: A preliminary market map 

We have seen a proliferation of promising AI-enabled vertical software products in a handful of sectors, and we are proud to be the earliest supporters of teams like Expedock, Pearl Street, Gryps, Hazel, and Federato.  

Many initial intelligently automated vertical software products target the largest sectors of the economy (we’ve looked separately at the ecosystem of AI in healthcare companies here). We’ll update this market map over time, and we’re eager to include companies unlocking automation potential in industries that have seen fewer capable purpose-built tools in the past.

What we are looking for 

We hope to support many more founders delivering on a new and bigger promise of vertical software. Standout teams that we currently support – or just simply admire – typically excel on a few dimensions: 

  1. They have an unfair data advantage.
  2. They have identified substantial automation potential. 
  3. They can communicate their value without invoking AI. 

We want to hear from you 

If you share our conviction that we’re entering a new golden age of vertical software and you’re exploring startup ideas that help expert operators streamline and grow their businesses through intelligent automation, we would love to hear from you. Reach out at if you’re working on something impactful.

Perspectives in AI: From LLMs to Reasoning with Edward Hu, Inventor of LoRA and μTransfer

I recently hosted a fireside chat with AI researcher Edward Hu. Our conversation covered various aspects of AI technology, with a focus on two key inventions Edward Hu pioneered: Low Rank Adaptation (LoRA) and μTransfer, which have had wide ranging impact on the efficiency and adoption of Large Language Models. For those who couldn’t attend in person, here is a recap (edited and summarized for length).

Aparna:  Welcome, everyone to the next edition of the ‘Perspectives on AI’ fireside chat series at Pear VC. I’m Aparna Sinha, a partner at Pear VC focusing on AI, developer tooling and cloud infrastructure investments. I’m very pleased to welcome Edward Hu today. 

Edward is an AI researcher currently at Mila in Montreal, Canada. He is pursuing his PhD under Yoshua Bengio, who is a Turing award winner. Edward has a number of inventions to his name that have impacted the AI technology that you and I use every day. He is the inventor of Low Rank Adaptation (LoRA) as well as μTransfer, and he is working on the next generation of AI reasoning systems. Edward, you’ve had such an amazing impact on the field.  Can you tell us a little bit about yourself and how you got started working in this field? 

Edward: Hello, everyone. Super happy to be here. Growing up I was really interested in computers and communication. I decided to study both computer science and linguistics in college. I got an opportunity to do research at Johns Hopkins on empirical NLP, building systems that would understand documents, for example. The approach in 2017, was mostly building pipelines. So you have your name entity recognition module, that feeds into maybe a retrieval system, and then the whole thing in the end, gives you a summarization through a separate summarization module. This was before large language models. 

I remember the day GPT-2 came out. We had a lab meeting and everybody was talking about how it was the same approach as GPT, but scaled to a larger data set and a larger model. Even though it was less technically interesting, the model was performing much better. I realized there is a limit to the gain we have from engineering traditional NLP pipelines. In just a few years we saw a transition from these pipelines to a big model, trained on general domain data and fine tuned on specific data. So when I was admitted as an AI resident at Microsoft Research, I pivoted to work on deep learning. I was blessed with many mentors while I was there, including Greg Yang, who recently started xAI. We worked on the science and practice of training huge models and that led to LoRA and μTransfer.

More recently, I’m back to discovering the next principles for intelligence. I believe we can gain much capability by organizing computation in our models. Is our model really thinking the way we think? This motivated my current research at Mila on robust reasoning.

Aparna: That’s amazing. So what is low rank adaptation in simple terms and what is it being used for? 

Edward: Low Rank Adaptation (often referred to as LoRA) is a method used to adapt large, pre-trained models to specific tasks or domains without significant retraining. The concept is to have a smaller module that contains enough domain-specific information, which can be appended to the larger model. This allows for quick adaptability without altering the large model’s architecture or the need for extensive retraining. It performs as if you have fine tuned a large model on a downstream task.

For instance, in the context of diffusion models, LoRA enables the quick adaptation of a model to particular characters or styles of art. This smaller module can be quickly swapped out, changing the style of art without major adjustments to the diffusion model itself.

Similarly, in language processing, a LoRA module can contain domain-specific information in the range of tens to hundreds of megabytes, but when added to a large language model of tens of gigabytes or even terabytes, it enables the model to work with specialized knowledge. LoRA’s implementation allows for the injection of domain-specific knowledge into a larger model, granting it the ability to understand and process information within a specific field without significant alteration to the core model.

Aparna: Low rank adaptation seems like a compelling solution to the challenges of scalability and domain specificity in artificial intelligence. What is the underlying principle that enables its efficacy, and what led you to develop LoRA?

Edward: We came up with LoRA two years ago, and it has gained attention more recently due to its growing applications. Essentially, LoRA uses the concept of low rank approximation in linear algebra to create a smaller, adaptable module.This module can be integrated into larger models to customize them towards a particular task.

I would like to delve into the genesis of LoRA. During my time at Microsoft, when GPT-3 was released and the OpenAI-Microsoft partnership began, we had the opportunity to work with the 175-billion-parameter model, an unprecedented scale at that time. Running this model on production infrastructure was indeed painful.

Firstly, without fine-tuning, the model wasn’t up to our standards. Fine-tuning, is essential to adapt our models to specific tasks, and it became apparent that few-shot learning didn’t provide the desired performance for a product. Although once fine-tuned, the performance was amazing, the process itself was extremely expensive.

To elucidate, it required at least 96 Nvidia V100s, which was cutting-edge technology at the time and very hard to come by, to start the training process with a small batch size, which was far from optimal. Furthermore, every checkpoint saved was a terabyte in size, which meant that the storage cost was non-negligible, even compared to the GPUs’ cost. The challenges did not end there. Deploying the model into a product presented additional hurdles. If you wanted to customize per user, you had to switch models, a process that took about a minute with such large checkpoints. The process was super network-intensive, super I/O-intensive, and simply too slow to be practical.

Under this pressure, we sought ways to make the model suitable for our production environment. We experimented with many existing approaches from academia, such as adapters and prefix tuning. However, they all had shortcomings. With adapters, the added extra layers led to significant latency, a nontrivial concern given the scale of 175 billion parameters. For prefix tuning and other methods, the issue was performance, as they were not on par with full fine-tuning. This led us to think creatively about other solutions, and ultimately to the development of LoRA.

Aparna: That sounds like a big scaling problem, one that must have prevented LLMs from becoming real products for millions of users. 

Edward: Yes, I’ll proceed to elaborate on how we solved these challenges, and I will discuss some of the core functionalities and innovations behind LoRA.

Our exploration with LoRA led to impressive efficiencies. We successfully devised a setup that could handle a 175 billion parameter model. By fine-tuning and adapting it, we managed to cut the resource usage down to just 24 V100s. This was a significant milestone for our team, given the size of the model. This newfound efficiency enabled us to work with multiple models concurrently, test numerous hyperparameter combinations, and conduct extensive model trimming.

What further enhanced our production capabilities was the reduction in checkpoint sizes, from 1 TB to just 200 megabytes. This size reduction opened the door to innovative engineering approaches such as caching in VRAM or RAM and swapping them on demand, something that would have been impossible with 1 TB checkpoints. The ability to switch models swiftly improved user experience considerably.

LoRA’s primary benefits in a production environment lie in the zero inference latency, acceleration of training, and lowering the barrier to entry by decreasing the number of GPUs required. The base model remains the same, but the adaptive part is faster and smaller, making it quicker to switch. Another crucial advantage is the reduction in storage costs, which we estimated to be a reduction by a factor of 1000 to 5000, a significant saving for our team.

Aparna: That’s a substantial achievement, Edward, paving the way for many new use cases.

Edward: Indeed. Now, let’s delve into how LoRA works, particularly for those new to the concept.  LoRA starts with fine-tuning and generalizes in two directions. The first direction concerns which parameters of the neural network – made up of numerous layers of weights and biases – we should adapt. This could involve updating every other layer, every third layer, or specific types of layers such as the attention layers or the MLP layers for a transformer.

The second direction involves the expressiveness of these adaptations or updates. Using linear algebra, we know that matrices, which most of the weights are, have something called rank. The lower the rank, the less expressive it is, providing a sort of tuning knob for these updates’ expressiveness. Of course, there’s a trade-off here – the more expressive the update, the more expensive it is, and vice versa.

Considering these two directions, we essentially have a 2D plane to help navigate our model adaptations. The y-axis represents the parameters we’re updating – from all parameters to none, which would retain the original model. The parameters of our model exist on a plane where the x-axis signifies whether we perform full rank updates or low rank updates. A zero rank update would equate to no updating at all. The original model can be seen as the origin, and fine tuning as the upper right corner, indicating that we update all parameters, and these updates are full rank.

The introduction of LoRA allows for a model to move freely across this plane. Although it doesn’t make sense to move outside this box, any location inside represents a LoRA configuration. A surprising finding from our research showed that a point close to the origin, where only a small subset of parameters are updated using very low rank, can perform almost as well as full fine tuning in large models like GPT-3. This has significantly reduced costs while maintaining performance.

Aparna: This breakthrough is not only significant for the field as a whole, but particularly for OpenAI and Microsoft. It has greatly expanded the effectiveness and efficiency of large language models.

Edward: Absolutely, it is a significant leap for the field. However, it’s also built on a wealth of preceding research. Concepts like Adapters, Prefix Tuning, and the like have been proposed years before LoRA. Each new development stands on the shoulders of prior ones. We’ve built on these works, and in turn, future researchers will build upon LoRA. We will certainly have better methods in the future.

Aparna: From my understanding, LoRA is already widely used. While initially conceived for text-based models, it’s been applied to diffusion models, among other things.

Edward: Indeed, the beauty of this approach is its general applicability. Whether deciding which layers to adapt or how expressive the updates should be, these considerations apply to virtually any model that incorporates multiple layers and matrices, which is characteristic of modern deep learning. By asking these two questions, you can identify the ideal location within this ‘box’ for your model. While a worst case scenario would have you close to the upper right, thereby not saving as much, many models have proven to perform well even when situated close to the lower left corner. LoRA is also supported in HuggingFace nowadays, so it’s relatively easy to use. 

Aparna: Do you foresee any potential challenges or limitations in its implementation? Are there any other domains or innovative applications where you envision LoRA making a significant impact in the near future?

Edward: While LoRA presents exciting opportunities, it also comes with certain challenges. Implementing low rank adaptation requires precision in crafting the smaller module, ensuring it aligns with the larger model’s structure and objectives. An imprecise implementation could lead to inefficiencies or suboptimal performance. Furthermore, adapting to rapidly changing domains or highly specialized fields may pose additional complexities.

As for innovative applications, I envision LoRA being utilized in areas beyond visual arts and language. It could be applied in personalized healthcare, where specific patient data can be integrated into broader medical models. Additionally, it might find applications in real-time adaptation for robotics or enhancing virtual reality experiences through customizable modules.

In conclusion, while LoRA promises significant advancements in the field of AI, it also invites careful consideration of its limitations and potentials. Its success will depend on continued research, collaboration, and innovative thinking.

Aparna: For many of our founders, the ability to efficiently fine tune models and customize them according to their company’s unique personality or data is fundamental to constructing a moat. What your work has done is optimize this process through tools like Lora and μTransfer. Would you tell us now about μTransfer, the project you embarked upon post your collaboration with Greg Yang on the theory of infinity with neural networks.

Edward: The inception of μTransfer emerged from a theoretical proposition. The community has observed that the performance of a neural network seemed to improve with its size. This naturally kindled the theoretical question, “What happens when the neural network is infinitely large?” If one extrapolates the notion that larger networks perform better, it stands to reason that an infinitely large network would exhibit exceptional performance. This, however, is not a vacuous question.

When one postulates an infinite size, or more specifically, infinite width for a neural network, it becomes a theoretical object open to analysis. The intuition being, when you are summing over infinitely many things, mathematical tools such as convergence of random variables come into play. They can assist in reasoning about the behavior of the network. It is from this line of thought that μTransfer was conceived. In essence, it not only has practical applications but is also a satisfying instance of theory and empirical applications intersecting, where theory can meaningfully influence our practical approaches.

I’d like to touch upon the topic of hyperparameter training. Training large AI models often involves significant investments in terms of money and compute resources. For instance, the resources required to train a model the size of GPT-3 or GPT-4 are substantial. However, a frequently overlooked aspect due to its uncertainty is hyperparameter tuning. Hyperparameters are akin to knobs or magic numbers that need to be optimized for the model to train efficiently and yield acceptable results. They include factors like learning rate, optimizer hyperparameters, and several others. While a portion of the optimal settings for these has been determined by the community through trial and error, they remain highly sensitive. When training on a new dataset or with a novel model architecture, this tuning becomes essential yet again, often involving considerable guesswork. It turns out to be a significant hidden cost and a source of uncertainty.

To further expound on this, when investing tens of millions of dollars to train the next larger model, there’s an inherent risk of the process failing midway due to suboptimal hyperparameters, leading to a need to restart, which can be prohibitively expensive. To mitigate this, in our work with μTransfer, we adopt an alternative approach. Instead of experimenting with different hyperparameter combinations on a 100 billion parameter model, we employ our method to reduce the size of the model, making it more manageable.

In the past, determining the correct hyperparameters and setup was akin to building proprietary knowledge, as companies would invest significant time experimenting with different combinations. When you publish a research paper, you typically disclose your experimental results, but rarely do you share the precise recipe for training those models. The working hyperparameters were a part of the secret. However, with tools like μTransfer, the cost of hyperparameter tuning is vastly reduced, and more people can build a recipe to train a large model.

We’ve discovered a way to describe a neural network that allows for the maximal update of all parameters, thus enabling feature learning in the infinite-width limit. This in turn gives us the ability to transfer hyperparameters, a concept that might need some elucidation. Essentially, we make the optimal hyperparameters the same for the large model and the small model, making the transfer process rather straightforward – it’s as simple as a ‘copy and paste’.

When you parameterize a neural network using the standard method in PyTorch, as a practitioner, you’d observe that the optimal learning rate changes and requires adaptation. However, with our method of maximal update parameterization, we achieve a natural alignment. This negates the need to tune your large model because it will have the same optimal hyperparameters as a small model, a principle we’ve dubbed ‘mu transfer’. Indeed, “μ” in “μTransfer” stands for “maximal update,” which is derived from a parameterization we’ve dubbed “maximal update parameterization”.

To address potential prerequisites for this transfer process, for the most part, if you’re dealing with a large model, like a large transformer, and you are shrinking it down to a smaller size, there aren’t many restrictions. There are a few technical caveats; for instance, we don’t transfer regularization hyperparameters because they are more of an artifact encountered when we don’t have enough data, which is usually not an issue when pretraining a large model on the Internet.

Nonetheless, this transfer needs to occur between two models of the same architecture. For example, if we have GPT3 175 B for which we want to find the hyperparameters, we would shrink it down to GPT3 10 mil or 100 mil to facilitate the transfer of hyperparameters from the small model to the large model. It doesn’t apply to transferring hyperparameters between different types of models, like from a diffusion model to GPT.

Aparna: A trend in recent research indicates that the cost of training foundational models is consistently decreasing. For instance, training and optimizing a model at a smaller scale and then transferring these adjustments to a larger scale significantly reduces time and cost. Consequently, these models become more accessible, enabling entrepreneurs to utilize them and fine-tune them for various applications. Edward, do you see this continuing? 

Edward: Techniques like μTransfer, which significantly lower the barrier to entry for training large models, will play a pivotal role in democratizing access to these large models. For example, I find it particularly gratifying to see our work being used in the scaling of large language models, such as the open-source Cerebras-GPT, which comprises around 13 billion parameters or more. 

In our experiments, we found that using μTransfer led to superior hyperparameters compared to those discovered through heuristics in the GPT-3 paper. The improved hyperparameters allowed a 6.7 billion parameter model to roughly match the performance of a 13 billion parameter model, effectively doubling the value of the original model with only a 7% increase in the pre-training cost.

Aparna:   It appears that the direction of this technology is moving towards a world where numerous AI models exist, no longer monopolized by one or two companies. How do you envision the utilization of these models evolving in the next one or two years?

Edward: It’s crucial to comprehend the diverse ways in which computational resources are utilized in training AI models. To begin with, one could train a large-scale model on general domain data, such as the Pile or a proprietary combination of internet data. Despite being costly, this is typically a one-time investment, except for occasional updates when new data emerges or a significant breakthrough changes the model architecture.

Secondly, we have domain-specific training, where a general-purpose model is fine-tuned to suit a particular field like law or finance. This form of training doesn’t require massive amounts of data and, with parameter-efficient fine-tuning methods like LoRA, the associated costs are dropping significantly.

Finally, there’s the constant use of hardware and compute in inference, which, unlike the first two, is an ongoing cost. This cost may end up dominating if the model or domain isn’t changed frequently.

Aparna: Thank you for the comprehensive explanation. Shifting gears a bit, I want to delve into your academic pursuits. Despite your significant contributions that have been commercialized, you remain an academic at heart, now back at Mila focusing on your research. I’m curious about your perspectives on academia, the aspects of research that excite you, and what you perceive to be the emerging horizons in this space.

Edward: This question resonates deeply with me. Even when I was at Microsoft, amidst exciting projects and the training of large models, I would often contemplate the next significant advancements in the principles and fundamentals underpinning the training of these models. There are myriad problems yet to be solved.

Data consumption and computational requirements present unique challenges to current AI models like GPT-4. As these models are trained on increasingly larger data sets, we might reach a point where we exhaust high-quality internet content. Moreover, despite their vast data processing, these models fail at executing relatively simple tasks, such as summing a long string of numbers, which illustrates the gap between our current AI and achieving Artificial General Intelligence (AGI). AGI should be able to accomplish simple arithmetic effortlessly. This gap is part of what motivates my research into better ways to structure computation and enhance reasoning capabilities within AI.

Shifting back to the topic of reasoning, it’s an exciting direction since it complements the scaling process and is even enabled by it. The fundamental question driving our research is, “How can we convert computations, or flops, into intelligence?” In the past, AI was not particularly efficient at transforming compute into intelligence, primarily due to limited computational resources and ineffective methods. Although we’re doing a better job now, there’s still room for improvement.

The key to turning flops into intelligence lies in the ability to perform effective search processes. Intelligence, at its core, represents the capability to search for reasons, explanations, and sequences of actions. For instance, when devising a move in chess, one examines multiple possible outcomes and consequences—a form of search. This concept is not exclusive to games like chess but applies to any context requiring logical reasoning.

Traditional AI—often referred to in research communities as “good old fashioned AI” or “GOFAI”—performed these search processes directly in the solution space. It’s analogous to playing chess by examining each possible move directly. However, the efficiency of these processes was often lacking, which leads us to the development of modern methods.

The fundamental challenge we face in computational problem-solving, such as in a game of chess, is that directly searching the solution space for our next move can be prohibitively expensive, even when we try to exhaustively simulate all possibilities. This issue escalates when we extend it to complex domains like language processing, planning, or autonomous driving.

Today, deep learning has provided us with an effective alternative. Although deep learning is still a form of search, we are now exploring in the space of neural network weights, rather than directly in the solution space. Training a neural network essentially involves moving within a vast space of billions of parameters and attempting to locate an optimal combination. While this might seem like trading one immense search space for another, the introduction of optimization techniques such as gradient descent has made this search more purposeful and guided.

However, when humans think, we are not merely searching in the weight space. We are also probing what we might call the ‘concept space.’ This space consists of explanations and abstract representations; we formulate narratives around the entities involved and their relationships. Therefore, the next frontier of AI research, which we are currently exploring at Mila with Yoshua, involves constructing models capable of searching this ‘concept space.’

Building on the foundations of large-scale, deep learning neural networks, we aim to create models that can autonomously discover concepts and their relationships. This approach harkens back to the era of ‘good old fashioned AI’ where researchers would manually construct knowledge graphs and scene graphs. However, the major difference lies in the model’s ability to learn these representations organically, without explicit instruction.

We believe that this new dimension of search will lead to better ‘sample complexity,’ meaning that the models would require less training data. Moreover, because these models have a more structured, lower-dimensional concept space, they are expected to generalize much better to unseen data. Essentially, after seeing a few examples, these models would ideally know how to answer the same type of question on unseen examples.

Aparna: Thank you, Edward. Your insights have been both practical, pertaining to present technologies that our founders can utilize, as well as forward-looking, providing a glimpse into the ongoing research that is shaping the future of artificial intelligence. Thank you so much for taking us through your inventions and making this information so accessible to our audience.

Join me for the next Perspectives in AI fireside, hosted monthly at Pear for up to date technical deep dives on emerging areas in Artificial Intelligence. You can find an archive of previous talks here.

Generative AI Tech Stack

We recently launched a dedicated AI track to the PearX program and have received a great response. Founders often ask us for guidance on how to build a moat for their AI startup. There are many aspects to this question but to kick things off, we are sharing a presentation I gave at SF Tech Week that covers background on the emergence of Generative AI, the highest priority areas of application particularly in enterprises, and what we believe enables a ‘moat’ for AI startups.

Generative AI Tech Stack Presentation at SF Tech Week

Generative AI is a game changing technology for humanity. A quote from one of my heroes, Professor Fei Fei Li at Stanford, and also was head of AI at Google Cloud for a while captures the excitement well:

“Endowing machines with generative capabilities, has been a dream for many generations of AI scientists” 

Seminal technologies which have led to the recent Generative AI breakthroughs include Cloud computing and within that advancements in GPUs, Kubernetes and open source frameworks like PyTorch provide an efficient and widely accessible substrate for model training and inference.

Research breakthroughs on the transformer neural network, its use on internet scale datasets and recent advancements in AI alignment are at the heart of most of the Generative AI capabilities today.

By no means are we at a peak yet, as research continues to improve efficiency at the hardware, software and services layers. Most interestingly to increase context lengths and optimize AI application architectures for accuracy, latency, and reliability. We cover some of these topics in depth in our Perspectives on AI fireside series.

It is clear that Generative AI techniques apply to multiple modalities. There has been a steady stream of models, both open source and proprietary in the major areas of NLP, Image, Video, Voice and also physical synthesis of Proteins.

Applications to both consumer and enterprise software abound and are already starting to change the shape of what software can do. We highlight some of the opportunities to build vertical and horizontal applications as well as tooling and infrastructure.

Of course there is hype when it comes to Generative AI, and in some sense it is almost too easy to create new functionality by building a thin layer over a foundation model that somebody else has built. While there are some businesses to be built in that way, for a venture scale business, we posit that a deeper moat is required to build. A large business that benefits rather than crumbles from rapid evolution of technology at the lower layers of the AI stack requires several moats.

Our thesis is that Applications will be composed of ensembles of specialized models, not just foundation models, but specialized models that are customized via fine tuning or in-context learning or a range of other techniques to complete part of a use case or workflow. These specialized models should utilize proprietary data specific to a domain and help to personalize the output of the application as well as ensure accuracy. A by product may also be lower cost to serve. Overall such an architecture will be a way to build lasting value and be more immune to disruption.

Tooling and infrastructure supporting the development of new applications of this kind is second part of our investment thesis. In particular, data and tooling companies to evaluate and ensure safety, accuracy, and privacy of these applications will be in demand. Lastly a few new infrastructure companies and capabilities will advance the development of these applications. We see emerging companies at every layer of the AI stack (slide 9). With that thesis in mind,  building a moat is fundamentally not that different in AI than in any other emerging space (slide 10).

Enterprise readiness for adoption of AI is arguably higher than it has ever been with the widespread acceptance of cloud computing, API integrations, and existing investments in data analytics teams and software. The hurdles to enterprise adoption are also not new, these are the same requirements that any cloud service has to meet, with perhaps a stronger need for ease of use and simplicity given the lack of existing AI/ML talent.

We conclude by quoting what many others have already said, that this is a great time to start a company!

Unleashing the power of Generative AI: an invitation to ambitious founders

We kicked off this week with the announcement of our 4th seed stage fund, one of the largest of its kind, raising $432M to seed the next generation of startups. Today we are thrilled to announce a dedicated Artificial Intelligence startup package, PearX for AI, which offers each founding team $250K in cloud credits, access to beta APIs, expanded check sizes up to $5M, 1:1 expert technical advice, customer introductions, AI talent matching and a curated community of AI practitioners who connect and learn from each other. Applications are open now through June 10th.

PearX for AI, is dedicated to the most ambitious technical founders, interested in building groundbreaking applications, tooling and infrastructure to power the Artificial Intelligence driven revolution. This program provides the resources, expertise, and customer connections you need to build the future. 

Who should apply?

PearX for AI will be a small tight-knit community with up to 10 startups selected into our inaugural summer batch. Pear specializes in working deeply with our portfolio companies to solve the most challenging problems startups face during the 0 to 1 journey from idea to product market fit. No idea is too early or too controversial. The program requires technical depth with a focus in AI, and the application process is tailored towards CS / Engineering graduates, PhDs, researchers and other technical professionals who have built AI driven applications in the past. We look for a combination of market knowledge, technical strategy and coding skills. The program will further build upon these skills and help round out your team’s capabilities in any areas that may need support. It will also connect you with Pear’s community of AI entrepreneurs.

Artificial Intelligence is a horizontal technology with the potential to impact many industries. We believe generative AI is a game-changer for consumer, social and enterprise applications. Particularly Healthcare, Legal, Financial Services, Retail, Logistics, Fashion, Design, Media, Gaming, Manufacturing, Energy, Industrial and Biotechnology to name a few verticals ripe for AI innovations. Pear’s AI team is especially skilled in enterprise AI adoption and will work with exceptional founders to craft solutions for the most pressing enterprise needs. 

We are deep technologists ourselves and value founders working on next generation Natural Language Understanding, Image generation, Computer Vision, Protein and Molecular synthesis and Robotics and Simulation technology. We have experts focused on applications of AI to developer tooling, open source software, sales, support, HR, R&D, design and education. We are bullish on infrastructure software, data services, and tools that reimagine the tech-stack for optimal AI performance.

What benefits will I receive?

Pear’s AI track comes with $250K in cloud credits, early access to new APIs and models, technical support from practitioners, mentorship from specialist AI experts, enterprise customer introductions, access to a talented and like-minded community, and an extended virtual cohort of AI founders. Pear will also extend larger check sizes for AI startups that require additional upfront investment, and provide introductions to strategic angel investors. Finally Pear will prepare all founders in the program for Series A fundraising. Our programs have an 87% success rate for series A and beyond investment by top tier funds.

Why now?

We’re living in an era of unprecedented technological advancement. AI is re-shaping our present and enabling some of the most significant breakthroughs of our lifetimes. The pace of this technological shift is breathtaking. The last time we witnessed a transformation of this nature was in 2000 with web technologies. There was a bubble then, just as there is significant hype around Generative AI now, but from it emerged breakout companies like Google, VMware, Salesforce and more. Breakout companies will be built now in this similar environment.

What does this mean for founders? Opportunity! There has never been a better time to start a company, but navigating the hype and shifting landscape that surrounds technical breakthroughs that are still in progress, requires expertise, judgment, and partners who will tell you hard truths and stand steady through difficult times, helping you secure the resources required – both financial and human.

Why Pear?

We have been investing in AI startups for several years and have a portfolio of AI powered companies that span verticals including consumer social, gaming, retail, healthcare, fintech, insurance, property technology, infrastructure, databases, deep tech, and more. 

With PearX for AI, we have pulled together the resources, community, and expertise to help founders discern signal from noise and succeed in building industry defining companies using the latest breakthroughs in Generative AI. The first track of PearX for AI is set to start in July, with a small cohort of founders who have a proven track record in applying AI to real-world problems. If you’re ready to take on this challenge and shape the future with AI, we invite you to apply now here. Let’s build the future together!

AR/VR/XR/PEAR: our call for mixed reality builders 

In the past few months, we have seen the beginnings of rising interest in building AR, VR, mixed reality, and extended reality infrastructure and applications (we’ll just call it “XR” for short): XR applications to PearX are up 5x this year, dedicated XR hacker groups are proliferating at top engineering schools like Harvard and Stanford, and our tracking shows hundreds of founders have left venture-backed and major tech companies to build in the XR space.  

We think this is because XR has the potential to represent one of the most consequential new platforms of the coming decade, and there is substantial alpha to be had for builders who study this burgeoning space and seize early opportunities. 

We expect interest in building in XR only to increase dramatically – particularly following Apple’s upcoming Reality Pro headset announcement. We see builders with a measured sense of urgency having an advantage, and we’ve put together a high-level guide for exploring ideas in XR. What follows is merely one way of cataloging opportunities; we would love to meet and speak with founders who are building early, quickly, and curiously in the broader XR space. 

XR Builder’s Guide to Exploring Ideas

A Familiar “Infrastructure and Applications” Approach

With any new technology, there are opportunities in foundational infrastructure (making the technology easier to deploy or adding capabilities to what can be built for / with it) and novel applications (tools built with the new technology to help users achieve something they could not previously do).

This approach often starts by asking what new infrastructure developers are building, and then asking what applications can be built atop it. 

In XR, substantial existing infrastructure will be first-party specific to headset makers. So, it is worth considering what initial applications may be built on foundations purpose-built for available devices – and which use cases may find breakout success among users of those devices. 

XR applications


Gaming appears to be the first breaking wave of consumer XR, and will likely lead the way for the foreseeable future. Unity’s 2023 report showed that more than half of all console developers are now building for VR platforms, too. It’s been said that “Every game developer wants to be a VR game developer – they just need to find a way to get paid for it.” This may not be a problem soon enough.

According to The Economist and Omdia, global gaming spending will eclipse $185B this year, with half of consumer spend going to mobile games. As AAA titles and boredom-busting mobile games alike are rendered in XR, it stands to reason that anyone with access to an XR device will prefer gameplay in a more immersive form – meaning that a sizable share of the gaming market may shift to XR in the next decade.

Gamified consumer training is already proving its effectiveness in athletics: thousands of baseball players, amateur and professional alike – including the reigning MVP – use WIN Reality’s Quest app to perfect their swings, with measurable improvements on their performance. 

We are also excited about consumer applications in more passive streaming entertainment, social content sharing, education, and commerce. Many of the hours of daily screen time spent watching asynchronous content – social media feeds or professional productions – or browsing e-commerce sites may feel more vivid in an immersive environment. 


Hardware fragility and cost may prevent near-term widespread enterprise adoption of B2B XR across business types. Meanwhile, few people – including us – are excited about open-ended co-working in an immersive XR environment. 

But, there are impactful vertical and horizontal applications alike that may soon make enterprise XR familiar and essential in many use cases, especially at large companies. Horizontal enterprise tools may include general-purpose training and demo environments: collaborative spaces built to allow anyone to host a classroom or sales presentation. Early deployments of immersive training tools have shown efficacy in use cases as diverse as nursing, UPS driver safety, and law enforcement.    

For more specialized B2B applications, initial verticals may include architecture and interior design, product design and mechanical engineering, repair and maintenance diagnostics and installation, and healthcare diagnostics and treatment simulation – among many other sectors.

Key questions for XR application builders 

With any application, we encourage prospective founders to consider first-party risk: What core applications are platform creators likely to want to own? FaceTimeXR may be a more likely winner than a new developer’s video chat app for Apple RealityPro. But, a social app to flip through your photo library immersively, in real-time and alongside the friends in those images, may be less likely in Apple’s first-party domain.  

We also encourage XR builders to have an airtight “why XR” case for any application: what vital benefit does an immersive, co-present environment offer to your user over alternative interfaces? 

XR Infrastructure

Developer tools 

Wise founders will study the initial wave of XR adoption and ask which high-use applications are breaking, underwhelming, or borderline impossible on existing first-party or open infrastructure. Many of the most most compelling opportunities will be in bolstering core experiences: interactivity and copresence, audio, messaging, 3D asset generation, 2D/3D web interoperability, streaming reliability and observability. 

Monetization enablement 

An entire ecosystem of companies will support familiar business models, applied to XR use cases. While many elements of these business models may be unchanged for a new interface, there will undoubtedly be novel components. E-commerce checkout flows will feature transaction-splitting for social shopping. A wave of analytics and marketing tools will help businesses identify lucrative, impression-rich actions users take in XR applications, and usage-based billing providers will emerge to track and monetize novel ways of product use. 

Key questions for XR infrastructure builders 

Although removed from the end XR application consumer, XR infrastructure builders should start with this headset-adorned or AR-phone-app wielding user profile in mind. In a nascent space, an infrastructure builder needs to be sure there are enough end users who will in turn adopt what your own customers build with your tool or on your platform. Even if a developer can build something novel and powerful with your tools, your success relies on that developer’s distribution.   


These are merely a few of the possible areas to explore. The promise of XR lies in the experiences we cannot yet clearly see, and the infrastructure built to enable them. If you’re building now — send us a note to And if you’ll be in LA for LA Tech Week, join us on 6/7 at our XR breakfast — there will be plenty to discuss the week of Apple WWDC!