Frameshifts with Benjamin Arya

About Frameshifts with Benjamin Arya

Spotlighting the world’s most ambitious biotech founders, investors, researchers and change-makers. frameshifts.bio (https://frameshifts.bio?utm_medium=podcast)

Publisher

Benjamin Arya

Category

science · technology

Language

en

Explicit

No

First episode

9 Jan 2024

Latest episode

31 Aug 2026

Latest episodes

29 episodes in the feed.

1. 

   31 Aug 2026

   Lucas Mair on AI, Robotics, and the Future of Fully Automated Biology | Frameshifts Episode #15

   This week, my X feed exploded with news that Anthropic had released hardware standards that allow AI agents to operate lab equipment directly. Their new “Model Hardware Standard” or MHS, built in collaboration with the Howard Hughes Medical Institute, allows an LLM to operate microscopes, liquid handlers, robotic arms and dozens of other instruments that scientists use to conduct experiments and to build the future. Until now, there has been no standard way to connect an LLM to a physical instrument. Every time you want to connect two pieces of equipment together, you need to create a bespoke integration that requires some grad student to write custom software. And grad students are busy. So when a lab wants to increase throughput, the incentive structure is usually to just hire more grad students or pressure encourage everyone in the lab to work longer hours. And scientists already work much longer hours than other professions. The standard informal working schedule at many elite labs across the US and China is something close to 996, often for mediocre salaries in the range of $35-70k. Of course, most people know that it’s silly for humanity to send our best and brightest to spend the better half of a decade getting PhDs in advanced bioengineering just to spend most of their time as manual labour meat sacks moving microliters of liquids between plastic tubes by hand. The bottleneck in biology has always been the ability to perform rapid iteration experiments to collect real world data. Intellect has been abundant in science since the 1950s, and with the advent of super-intelligent models, it is already becoming commoditized. These days, what distinguishes Earth’s most impactful and highest throughput labs is largely prestige-driven meat sack talent acquisition, work ethic, pipetting technique and ability to fund expensive experiments. This week’s episode with Lucas Mair is all about what the world will look like when we realise that robots exist and start building accessible, intuitive software to allow them to perform closed-loop laboratory experiments. Lucas is a very German MD-PhD who took an unusual route into biology. Computer science first, then medicine, then neural engineering at TU Munich, and before any of that, building autonomous drone swarms for search and rescue. He thinks like a control theorist, and when a control theorist looks at how a modern biology lab actually runs, he doesn't see cutting-edge science. He sees a factory floor from 1975. Lucas will tell you that most labs already have an Opentrons or something similar gathering dust in the corner. My own lab had one that I have never seen anyone use. Grants pay for reagents and salaries, not for engineers to automate your protocols. Grad students are cheaper than robots, and publishing pressure means the three months you'd spend automating a workflow is three months you're not putting out papers, which is the equivalent of suicide by irrelevancy for many high-profile labs. Thus, the rational move, for every individual scientist and every individual lab, is to just keep pipetting. And so that’s what everyone does. Which is insane, because the robots already exist. We've had them for years. What we've been missing is the thing that turns "here's what I want to test" into machines actually doing it. The Anthropic news is a bit of a preview into what that might look like. But the reality of the current tech is a mixed picture. For example, scientists at Carnegie Mellon were able to use an AI agent with MHS to orchestrate a liquid handler, plate reader, robotic arm and monitoring cameras to get serial-dilution dose-response experiments done 3x faster (https://qz.com/anthropic-model-hardware-standard-ai-robots-lab-equipment-082826). Meanwhile at Genentech, humans had to intervene when Claude used MHS to coordinate a BCA protein assay (https://techstartups.com/2026/08/27/anthropic-launches-model-hardware-standard-to-let-ai-agents-control-physical-machines/) because the model mistook errors caused by bubbles for a software failure, and then responded in a way that produced yet more bubbles. Lucas Mair looks at biology as a control and engineering problem. You have a complex biological system, and you want to move it from State A to State B (e.g. diseased to healthy). But the problem is there is an unimaginably large search space of possibilities. For example, there are 2020 or 104,857,600,000,000,000,000,000,000 ways of making a 20 amino acid protein sequence. Most of these sequences are useless but a few of them might help make humans immortal. One more thing to understand before you listen to the full conversation… If you’ve spent much time at all on the tech bro side of X, you’ll know about the scaling laws of large language models. Roughly speaking, intelligence scales linearly with the log of compute (or data). Thus to improve your model, you need more data. But most of the recent improvement in LLM performance across disciplines hasn’t been in training larger language models. Rather, the most noticeable advances have been from “reinforcement learning from verifiable rewards” or RLVR. Basically a form of reinforcement learning where the reward you give the model can be checked objectively rather than judged subjectively by humans. After all, if you want to build super-intelligent systems, then at some point, even the smartest humans are going to be useless in teaching them anything. RLVR is really easy to do in fields like math and coding because the correct answer is already known and you can check if the program passed the tests. But how do you do this in biology? You can’t just read all the papers because most biological data isn’t published and the results of many peer-reviewed papers can’t be reproduced by other labs (Cc: the reproducibility crisis in science). You don’t need more text tokens. You need more ground truth data (i.e. the actual underlying molecular biology) aimed at the model’s blind spots. The jagged frontier of AI progress is more like a recursive Sierpiński triangle. Each field, subfield and sub-sub-field of human endeavor has their own jagged frontiers that would benefit from RLVR. If you can generate data aimed at places where the model is uncertain or wrong, using real experiments, the efficiency gain can be enormous. Now imagine running that loop in biology, on its own. The model would propose the hypothesis, plan the experiment, dispatch agents to run the lab equipment and collect real measurements. The data then would get processed to generate the next hypothesis. This is a system that learns by doing, using verifiable, falsifiable predictions. It’s the same technique that helped AI vibe coding go from being an unusable joke to the default way that 90% of today’s code is written, including at frontier labs. Once you collect enough data on the model’s biological blind spots, you cross some invisible threshold of reliability that makes your AI model generally useful in autonomously solving biological engineering problems, inventing drugs, running experiments, and, eventually, agentically building biotech companies. Closed-loop automated labs could put real experimental biology in the hands of far more people than the credentialed few who have obtained their pipette licenses. One day, you might be able to prompt a model to fix your body, cure your cold or build your dog a cancer vaccine. Or maybe you already can, but just haven’t tried. Watch on YouTube. Listen on Apple Podcasts or Spotify. GUEST INFORMATION: * Lucas Mair, Co-Founder & CTO, ProviGenAI (https://provigen.ai/) * ProviGenAI (https://provigen.ai/) * Model Hardware Standard (Anthropic) (https://www.anthropic.com/news/model-hardware-standard-research-preview) * X (Twitter) (https://x.com/OrthoDim) * LinkedIn (https://de.linkedin.com/in/lucas-mair) CONNECT WITH US: * Website (https://frameshifts.org/) * Substack (https://frameshifts.bio/) * YouTube (https://www.youtube.com/@Frameshifts) * X (Twitter) (https://x.com/frameshiftspod) * LinkedIn (https://www.linkedin.com/company/frameshifts/) Get full access to Frameshifts with Benjamin Arya at frameshifts.bio/subscribe (https://frameshifts.bio/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

2. 

   5 Jul 2026

   Logan Thrasher Collins: The Trojan Horse That Could Fix Gene Therapy | Frameshifts Episode #14

   On this show, we’ve talked a lot about gene editors. These are little molecular machines derived from or inspired by the Nobel-prize winning CRISPR system that put gene editing on the radar of the general public. Once these editors get into a cell, they can cut out DNA, change the DNA sequence, insert entire copies of genes or delete faulty genes that cause illness. Think of them as tools to cut, paste, backspace and overwrite the code of life. But the primary challenge with these gene editors, besides off-target edits and low-but-steadily-improving editing efficiencies, is that they need to get into the cell to be able to do anything. Getting gene editors into cells is an entire field of biomedical engineering called gene delivery. And in comparison to the rapidly flourishing field of gene editor tech dev, the field of gene delivery is both under-appreciated and relatively stagnant. And that’s a massive problem. Because without effective delivery vectors, gene editors are limited to a very small number of conditions for which our delivery vectors happen to be okay at. Current clinical successes are concentrated in situations where delivery is unusually favorable: * Liver diseases, where lipid nanoparticles can efficiently deliver cargo to hepatocytes after IV injection * Blood disorders and some hematological cancers, where cells can be removed from the patient, genetically modified ex vivo using lentiviruses and then re-injected into the patient (e.g. HSC therapy for sickle cell anemia or CAR-T therapy for leukemia) * Diseases that only require a very small percentage of cells to be edited in order to be functionally cured * A limited number of tissues, such as the retina or CNS where viral vectors can achieve therapeutically meaningful gene transfer That's the ceiling today. But if we crack delivery, the potential of this field is disgustingly profound. Gene Delivery Vector + Gene Editor = Gene Therapy If humanity can solve both sides of this equation, we can, literally, cure anything. You have a genetic disease? Here’s an injection that gives all of your cells the healthy copy of the corrupted gene you inherited from your parents. You’re prone to cardiovascular disease? Here’s an injection that gives you new human-designed molecular machines that clear out your clogged arteries. You have a family history of cancer? Here’s an injection that replaces your genome maintenance and DNA machinery with engineered supernatural versions that can protect your cells from cancer-causing mutations for at least 1000 years. Worried about keeping up in a world of artificial superintelligence? Here's an injection that carries the morphological instructions for a population of your cortical neurons to rewire themselves for compatibility with high-bandwidth brain-machine interfaces, connecting you directly into Dario or Sam’s frontier models. Now, I know a lot has to go right for humanity to achieve mastery over the code of life and transcend the limits imposed by “fit enough to have offspring” evolutionary biology. So I’ll get my head out of the clouds for a moment. Currently, the most clinically successful delivery vehicle in the emerging field of gene therapy is a family of viral variants that we call Adeno-Associated Viruses, or AAVs. AAVs are DNA viruses that scientists have hijacked to make little viral couriers that carry a working gene into your cells. Within this group are many serotypes and engineered capsid variants, each with tropism for specific organs and tissue types. Significantly oversimplifying: AAV9 can penetrate the brain, AAV8 is strongly liver-tropic and other capsids are tuned for tissues like skeletal muscle or retina. But somewhere between a third and two-thirds of people already have antibodies against AAVs, left over from ordinary childhood infections with the wild versions. If you’re one of them, a gene therapy that could save your life may simply not be available to you, because your immune system will neutralize the courier before it arrives. And even if you’re in the clear, most AAV therapies work exactly once. The first dose teaches your body to recognize the vector, so a second one gets destroyed the moment it touches your immune system. For a whole field built on delivering genes into people, the delivery itself is one of the deepest unsolved problems there is. Logan Thrasher Collins has spent his career on that problem. As a PhD student in biomedical engineering at Washington University in St. Louis, he invented a new gene-delivery modality called vaultAAV, which does something a bit strange: it hides the AAV inside a protein vault. Wow. Incredible. But what the hell is a protein vault? Vaults are odd, barrel-shaped organelles our own cells churn out in huge numbers, and nobody is entirely sure what they’re for (https://www.cellgs.com/blog/cell-vaults-the-next-big-thing-in-biology.html). What Logan realized is that because the body recognizes a protein vault as self, they act as immunologically invisible Trojan horses. Cells take them up, the vault disassembles in the acidifying endosome, and the AAV escapes to deliver its DNA to the nucleus, all without the immune system ever getting a look at the AAV courier hiding inside. In this episode, Logan walks us through his entire scientific journey, from designing an aggregating antimicrobial peptide as a high school science fair project (which won the top prize at ISEF and got a minor planet named after him), to the insight that connected the protein vault literature with the AAV field’s antibody problem. We discuss the full landscape of gene delivery: AAVs, adenoviruses, LNPs, virus-like particles, and why the field’s relationship with each vector is driven more by reputation and vibes than by rigorous analysis of dosing and safety data. Logan also explains why vaultAAVs may enable something no other approach can: redosability. Currently, AAV capsids are engineered with surface mutations to escape antibodies, but each new variant can only be used once before the immune system learns the new epitope. Vault-shielded AAVs, because their outer surface is self, could theoretically be administered multiple times, allowing lower doses, reduced toxicity and lower manufacturing costs. Logan has since taken the vaultAAV work out of the lab and into a company. He co-founded Cathedral Therapeutics (https://www.cathedraltx.com/), a gene-delivery engineering company, alongside Prof. David Curiel, MD, PhD (https://radonc.wustl.edu/people/david-t-curiel-md-phd/) at Washington University, to build exactly this kind of stealth delivery system. In this episode, we also get into: * How 30 to 60% of people are locked out of AAV gene therapy by the time they reach adulthood * How hiding a virus inside one of your own organelles gets it past the immune system * Why vaultAAVs don’t just dodge antibodies but also boost transduction up to 5x (in some cell lines) * Why gene therapies cost $850K to $3.5M a dose, and how synthetic biology could make them cheap * How a high-school science fair peptide led to Logan getting a minor planet named after him * How we can engineer adenoviruses to cross the blood-brain barrier in order to treat neurodegenerative disease and mental illnesses * What it means that our cells are full of a mysterious organelle nobody can fully explain Gene editing technologists like David Liu, Feng Zhang and my friends in the AbuGoot lab have spent two decades proving that CRISPR-based systems can fix faulty genes. The much harder problem is how we can get gene therapies into everyone who needs them, and eventually make them as safe, cheap and ubiquitous as Ozempic. To do that, we need to be able to inject them more than once, and get them to our target tissues without the immune system getting in the way. That’s the problem Logan is working on. Watch on YouTube, listen on Apple Podcasts or Spotify. GUEST INFORMATION: * Logan Thrasher Collins, PhD, Washington University in St. Louis (http://logancollinsblog.com) * Cathedral Therapeutics (https://www.cathedraltx.com/) * VaultAAV preprint (https://doi.org/10.1101/2023.11.29.569229) * Synthetic Biology for AAV Manufacturing (https://doi.org/10.1021/acssynbio.2c00589) * X (Twitter) (https://x.com/LoganTCollins/status/2067740366820786503) CONNECT WITH US: * Website (https://frameshifts.org/) * Substack (https://frameshifts.bio/) * YouTube (https://www.youtube.com/@Frameshifts) * X (Twitter) (https://x.com/frameshiftspod) * LinkedIn (https://www.linkedin.com/company/frameshifts/) * TikTok (https://www.tiktok.com/@frameshiftspod) Get full access to Frameshifts with Benjamin Arya at frameshifts.bio/subscribe (https://frameshifts.bio/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

3. 

   20 Jun 2026

   Joe Betts-LaCroix on Retro, Reprogramming, and the First Real Longevity Drugs | Frameshifts Episode #13

   Most longevity talk is still stuck within two bad schools of thought. On one side, weak-sauce incrementalism: supplements, vibes and tiny effect sizes. On the other, immortality rhetoric that asks you to throw out the data altogether. Joe Betts-LaCroix has very little patience for either. He’s the co-founder and CEO of Retro Biosciences, one of the most ambitious longevity companies in the world. Launched with an initial $180 million of capital from Sam Altman’s personal wealth, the company has since raised new round led by 4P Capital at a $1.8 billion valuation. Retro’s mission is to add ten healthy years to human life by going after the cellular mechanisms of aging directly. Betts-LaCroix is an unlikely person to be running a longevity company. He left school with a D average and spent six years in a shared house with a “bunch of musicians and artists and weirdos”, doing electronics work to pay rent. That somehow became biophysics at Harvard, MIT and Caltech, then building the world’s smallest personal computer (OQO, acquired by Google), and later founding an automated animal-research company (Vium, acquired by Recursion), before landing on the question he couldn’t walk away from, which is why we age and what we can actually do about it. A lot of our conversation is about what a serious longevity biotech company looks like once you strip the noise away. We dig into how Retro chose its three core programs, autophagy, partial reprogramming and blood stem cell replacement, out of an initial six. We talk about how the company walked away from their blood plasma program despite promising early data, and how their liver reprogramming results, dramatic as they were in mice, became commercially unviable once GLP-1 drugs reshaped the entire liver-disease market. Their first drug, a small-molecule autophagy therapy aimed at neurodegeneration, has since begun dosing patients in a Phase I trial in my home country of Australia. Then we go deep on GPT-4b micro, the protein-design model Retro built with OpenAI. Joe lays out what it actually is and isn’t, how it can rewrite up to 80% of a protein’s amino acids while still preserving fold and function, and why the redesigned Yamanaka factors it produced have already made their way into FDA submissions. It’s striking to me that this model (among several others in its class) keeps proposing designs no human would ever have come up with, and that general-purpose protein generation carries a biosafety tension Joe doesn’t try to dodge. From there, our conversation opens up into the bigger questions: why aging may not need to be formally recognized as a disease for meaningful therapies to win approval, what it takes to build a company around a problem this hard, and, near the end, what happens to human ambition and our sense of time when people stop living only for tomorrow. In this episode, we cover: * Why the target isn’t immortality but ten additional healthy years * Why one of its first clinical programs targets autophagy in neurodegeneration * How partial reprogramming evolved from whole-body proof of concept toward tissue- and cell-type-specific intervention * Why Retro walked away from a promising liver reprogramming program despite dramatic mouse data * How GPT-4b micro works as a protein-design model, and why it’s more than “an LLM for Yamanaka factors” * Why protein design may unlock improved versions of our own endogenous maintenance machinery * How Joe thinks about regulation, aging endpoints, and what will eventually force the FDA to adapt * Why meaningful missions tend to produce stronger companies and stronger teams * Why he sees longevity not as a billionaire side quest, but as one of the most important things a civilization could work on The most important problems in biology might also, if you’re careful about which ones you take on, turn out to be the solvable ones. GUEST INFORMATION: * Joe Betts-LaCroix, Co-Founder & CEO, Retro Biosciences (https://en.wikipedia.org/wiki/Joe_Betts-LaCroix) * Retro Biosciences (https://retro.bio/) * OpenAI x Retro Biosciences, GPT-4b micro (https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences/) * Defining a longevity biotechnology company (Nature Aging) (https://www.nature.com/articles/s41587-023-01854-0) * X (Twitter): @bettslacroix (https://x.com/bettslacroix) * LinkedIn (https://www.linkedin.com/in/bettslacroix) CONNECT WITH US: * Website (https://frameshifts.org/) * Substack (https://frameshifts.bio/) * YouTube (https://www.youtube.com/@Frameshifts) * X (Twitter) (https://x.com/frameshiftspod) * LinkedIn (https://www.linkedin.com/company/frameshifts/) * TikTok (https://www.tiktok.com/@frameshiftspod) Get full access to Frameshifts with Benjamin Arya at frameshifts.bio/subscribe (https://frameshifts.bio/subscribe?utm_medium=podcast&utm_campaign=CTA_4)

Reach and audience

Public platform figures. Ratings count people who left a rating, not total listeners.

Spotify

5.0 / 5

2 ratings

Contact Frameshifts with Benjamin Arya

Guest appearances

Books guests

Based on episode analysis; this does not confirm that the show is currently accepting guests.