Near Horizon

(2:05) LoRA origins — Edward describes the practical need that led to the creation of LoRA at Microsoft in 2021 while collaborating with OpenAI

(6:50) Why LoRA worked — Low-rank updates to large models made fine-tuning efficient and widely accessible

(9:45) Biological analogy — LoRA as the equivalent of learning a new instrument with shared prior knowledge

(11:35) Industry adoption — Edward reflects on companies thanking him for saving compute costs and resource usage

(13:30) Intelligence as compression — Discussion of whether brains and AI share a compression-based structure for learning

(17:15) Risks of over-biological inspiration — Why not everything in AI needs to imitate the brain

(18:45) MuTransfer explained — Training stability via infinite-width theoretical insights and their practical payoff

(23:00) Real-world impact of MuTransfer — Especially valuable for massive models at top labs like OpenAI, Anthropic

(26:30) The joy of theory with impact — The rare but powerful moment when theory turns into applied breakthrough

(28:05) Training trillion-parameter models — How MuTransfer enables precise hyperparameter prediction at scale

(30:30) Edward’s path — From Microsoft to PhD with Yoshua Bengio to OpenAI and now entrepreneurship

(34:00) Why leave OpenAI? — Personal itch for autonomy and new challenges, despite leaving pre-vesting

(37:20) From researcher to founder — Edward’s shift to launching multiple ventures with strong partners

(41:00) “Tech alone isn’t enough” — Why proprietary data + distribution is the real edge for AI startups

(44:00) Vertical vs horizontal plays — Studio model allows broad application of AI in many industries

(46:40) Laura-as-a-metaphor — Marc proposes a meta analogy: Edward as the foundation, fine-tuned by domain experts

(49:25) The fall of engineering costs — Why software is easier to build than ever, and what that unlocks

(52:00) The new venture model — VCs must adapt to a world with lower costs and smaller teams

(55:45) Edward’s self-awareness — On knowing his strengths and staying in his zone of genius

(57:10) The Compute Exchange — Edward unveils a new project: a dynamic GPU auction platform for buyers and sellers

(1:00:20) Compute is the new oil — Making GPU access as transparent and efficient as financial markets

(1:02:10) On IP and creator rights — Edward’s take on respecting creativity, compensating artists, and not undermining the creative industry

(1:06:00) The soul in human creation — Machines may be good imitators, but creativity may still be uniquely human

(1:08:00) Recommended Reading — Edward recommends The Creative Act: A Way of Being by Rick Rubin as a meditation on creativity across disciplines

(1:10:15) Outro — Marc reflects on Edward’s polymath career arc and cross-industry potential

(0:00) Introduction — Marc Jacobstein opens the episode and introduces Edward Hu, world-renowned AI researcher (LoRA, μTransfer) and now entrepreneur

(2:05) LoRA origins — Edward describes the practical need that led to the creation of LoRA at Microsoft in 2021 while collaborating with OpenAI

(6:50) Why LoRA worked — Low-rank updates to large models made fine-tuning efficient and widely accessible

(9:45) Biological analogy — LoRA as the equivalent of learning a new instrument with shared prior knowledge

(11:35) Industry adoption — Edward reflects on companies thanking him for saving compute costs and resource usage

(13:30) Intelligence as compression — Discussion of whether brains and AI share a compression-based structure for learning

(17:15) Risks of over-biological inspiration — Why not everything in AI needs to imitate the brain

(18:45) MuTransfer explained — Training stability via infinite-width theoretical insights and their practical payoff

(23:00) Real-world impact of MuTransfer — Especially valuable for massive models at top labs like OpenAI, Anthropic

(26:30) The joy of theory with impact — The rare but powerful moment when theory turns into applied breakthrough

(28:05) Training trillion-parameter models — How MuTransfer enables precise hyperparameter prediction at scale

(30:30) Edward’s path — From Microsoft to PhD with Yoshua Bengio to OpenAI and now entrepreneurship

(34:00) Why leave OpenAI? — Personal itch for autonomy and new challenges, despite leaving pre-vesting

(37:20) From researcher to founder — Edward’s shift to launching multiple ventures with strong partners

(41:00) “Tech alone isn’t enough” — Why proprietary data + distribution is the real edge for AI startups

(44:00) Vertical vs horizontal plays — Studio model allows broad application of AI in many industries

(46:40) LoRA-as-a-metaphor — Marc proposes a meta analogy: Edward as the foundation, fine-tuned by domain experts

(49:25) The fall of engineering costs — Why software is easier to build than ever, and what that unlocks

(52:00) The new venture model — VCs must adapt to a world with lower costs and smaller teams

(55:45) Edward’s self-awareness — On knowing his strengths and staying in his zone of genius

(57:10) The Compute Exchange — Edward unveils a new project: a dynamic GPU auction platform for buyers and sellers

(1:00:20) Compute is the new oil — Making GPU access as transparent and efficient as financial markets

(1:02:10) On IP and creator rights — Edward’s take on respecting creativity, compensating artists, and not undermining the creative industry

(1:06:00) The soul in human creation — Machines may be good imitators, but creativity may still be uniquely human

(1:08:00) Recommended Reading — Edward recommends The Creative Act: A Way of Being by Rick Rubin as a meditation on creativity across disciplines

(1:10:15) Outro — Marc reflects on Edward’s polymath career arc and cross-industry potential

(2:10) Frank's background – Stanford, Palantir, Google, Calico, now Stately Bio

(4:35) What Stately Bio does – Marrying brightfield imaging with modern ML for live cell analysis

(8:10) Why brightfield imaging matters – Cheap, live, centuries-old tech meets new machine learning

(12:05) Real-time monitoring and forecasting biological outcomes

(18:20) How living cell observation shrinks iteration cycles

(24:45) Applications to stem cell therapies: diabetes, Parkinson’s, liver, heart, kidney

(30:10) ML advantage – Machine vision finds subtle signals invisible to humans

(35:50) Stately Bio’s early success – New York Blood Center partnership

(40:25) The cost advantage of imaging vs. single-cell omics

(44:30) Platform vs pipeline – Why Stately is building therapies, not just a tool

(50:00) Future verticals: Biomanufacturing, synthetic biology applications

(54:50) Frank’s journey – From coder to founder to biotech CEO

(1:00:05) Challenges in filtering conflicting advice as a founder

(1:04:40) Surprise learnings from scaling Stately

(1:09:00) ISCT conference announcements – New breakthroughs in suspension cells imaging

(1:13:25) Big vision – Starting the phenomics revolution through live-cell analysis

Mark Jacobstein interviews Andrew Pannu, founding CEO of Sleuth Insights, a company providing AI-enabled competitive intelligence for BioPharma.

Guest: Andrew Penuuu, founding CEO of Sleuth Insights, previously an analyst at Deutsche Bank and COVID Pharma

OUTLINE:

0:00 - Introduction

1:47 - About Sleuth Insights

5:04 - Industry challenges in biotech intelligence

10:07 - Value of AI-powered analysis in biotech

13:45 - Product evolution and customer discovery

16:42 - Building a social media presence as distribution advantage

21:31 - Solo founder experience

26:30 - Evolution from self-serve software to "Swiss" model

32:45 - Vision for transforming BioPharma decision-making

35:26 - Reading recommendations