
AI x Bio | CEO at Ligo Biosciences | Prev. Medicine @UniofOxford
San Fransisco
Joined July 2024
- Tweets211
- Following1.3K
- Followers429
- Likes523
Ed Harris retweeted
The pharma folks are right to call out Dario for making bold claims about curing all disease.
They are also right that most necessary bio data does not yet exist.
Still, pharma has two crucial blindspots:
1) The industry has never witnessed what a small, well capitalized, and well incentivized silicon valley team can do (no instagram, whatsapp, ARM, cursor in pharma). They grossly underestimate what's coming for them.
2) They don't yet see just how much the labs *need* to win in bio to double TAM (there simply is not other accessible market), and just how much all of tech's liquidity relies on this happening.
Pharma is a risk management industry: you get fired for failing in old ways, not for failing in new ways. The industry will be shocked at how fast automated labs, better models, etc can be stood up by small, well capitalized teams that have only 1x to lose and thousands of x to gain.
Pharma doesn't quite yet have a traditional innovator's dillema. Their capital and GTM moats are still deep, but not for much longer (drugs going d2c, dev costs decreasing). Pharma's window to win is open, but it will close fast if they fail to take ownership of the incoming infrastructure explosion.
Ed Harris retweeted
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Ed Harris retweeted
This is a fantastic new BioAI benchmark that’s highly relevant to advancing the field! I just spoke with @ArdaGoreci, one of its co-founders, and they have even more exciting BioAI benchmarks in the pipeline!
Developing rigorous, biologically relevant benchmarks is absolutely critical to accelerating progress in BioAI!
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
Ed Harris retweeted
OpenAI built BioBasicsBench, a test that gives AI agents the messy raw data of real lab work, like gel photos and microscope videos, and grades whether they read it correctly.
GPT-6 Astra earned 66.7% of available credit on a new wet-lab benchmark, versus 43.8% for GPT-5.6 Sol.
BioBasicsBench, built with enzyme-design company Ligo, draws 100 tasks from scientists’ raw gels, colony plates, microscopy, sequencing reads and bench videos.
Its rubrics give partial credit, including for spotting errors or saying evidence can’t support conclusions.
At the xhigh setting, Astra used 4,704 output tokens per task, about half the older model’s, so longer reasoning didn’t drive the gain.
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
OpenAI's GPT-6 Astra is crushing older models on real-world biology tasks.
OpenAI just introduced BioBasicsBench, a new benchmark developed with biotech company Ligo.
It includes 100 tasks from real lab experiments, testing whether AI can:
> Analyze raw lab data
> Detect experimental errors
> Recognize unsupported conclusions
GPT-6 Astra and GPT-6.1 Sol are leading with scores around 60–65%, while older models are mostly below 50%.
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
Releasing some of our recent work with @OpenAI
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
4
26
1,517
Excited to release BioBasicsBench, a benchmark we developed with @OpenAI to test whether AI agents can interpret real lab data, spot experimental errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench reveals where models still struggle with routine experimental tasks and help measure whether they’re improving.
Explore the benchmark: openai.com/index/measuring-h…
A new way to measure how AI agents handle the routine analyses of wet-lab science—BioBasicsBench.
Developed with Ligo, a biotech company focused on enzyme design, the benchmark draws on 100 tasks from real experiments to test whether agents can interpret raw lab data, identify errors, and recognize when the evidence doesn’t support a conclusion.
Accelerating biological research starts with getting the everyday work right. BioBasicsBench shows where models need to improve, and helps us track what those gains mean for scientific progress, biosafety, and biosecurity.
openai.com/index/measuring-h…
Ed Harris retweeted
We're excited to announce Pathfinder's $16M in funding with our Series A led by Portage with participation from Reach Capital, Gradient and Stand Together Ventures Lab
If this animates and excites you, we're hiring: pathfindercard.com/join
Ed Harris retweeted
What?! 🤯 In the future, all surgeries will be performed by robots with AI agents! They will be much more precise and safe. Human surgeons may still be in the loop for some time, though.
“Last night an AI agent performed brain surgery on a mouse. We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorized surgery rig with cameras, and one goal: reduce seizures.”
Last night an AI agent performed brain surgery on a mouse.
We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures.
It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding.
A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent.
Thread on what it decided, what it saw, and the moment it refused to let me override it.
[Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...]
Ed Harris retweeted
Last night an AI agent performed brain surgery on a mouse.
We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures.
It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding.
A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent.
Thread on what it decided, what it saw, and the moment it refused to let me override it.
[Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...]
Ed Harris retweeted
Introducing Astraeus.
High mobility. Long duration operations. Greater freedom to maneuver.
Redefining what spacecraft can do in orbit.
Built by Orbital Operations.
Concept video produced by London Studio @ZeroOneCreative and a special thanks their team @ashkan01C, @Rupert01C and @james_01c
Ed Harris retweeted
ai x bio just had a ridiculous 48 hours - V2:
- Warp Sciences: AI x robotic labs for drug discovery
- Carbon-X: open model for DNA interpretation
- Nanograb: AI for in vivo targeted drug delivery
- Holon-S1: cancer biomarkers from pathology slides
- Pinal: protein design from plain language
- Vitrify: molecular simulations beyond DFT accuracy
- Biomni + Bayer: AI agent deployment
- Boltz + BMS: protein model deployment
- Atomwell: Bindwell's binding affinity model
- Omnii-1: AI-designed phage given to a patient
- Claude designed scorpion toxin binders
- OpenAI acquires Orchestra Bio
- Baker Lab: RFD4-Proteina
- Chaperone: agent framework for protein design
- BioEmu2: protein simulations
- Artios: enzyme design model
- OpenAI + PerturbAI: agents find brain disease clues
- Xaira X-Scientist
anything still missing?
Ed Harris retweeted
mRNA will be crucial to beating cancer and solving disease but only if we can deliver it to the right cells.
Nanograb's AI-designed multivalent nanoparticles address a cell's unique receptor fingerprint, now proven in vivo across multiple cell types and scaling to many more.
Ed Harris retweeted
ai x bio just had a ridiculous 48 hours:
- Omnii-1: AI-designed phage given to a patient
- Claude designed scorpion toxin binders
- OpenAI acquires Orchestra Bio
- Baker Lab: RFD4-Proteina
- Chaperone: agent framework for protein design
- BioEmu2: protein simulations
- Artios: enzyme design model
- OpenAI + PerturbAI: agents find brain disease clues
- Xaira X-Scientist
anything missing?
Ed Harris retweeted
These GOATS went open source!
Today we are excited to release Artios, a fully open source system for designing functional de novo enzymes.
Enzymes present a new frontier for de novo protein design models due to the challenge of coordinating precise chemistry, rather than binding a target.
Enzymes are life’s biological machines and by designing them we can truly start to engineer biology.
We believe these fundamental capabilities should be built in the open and shared with the community to progress science faster.
🧵 1/6
Ed Harris retweeted
Congratulations to Arda and his team on the release of Artios, an open-source BioAI model focused on designing enzymes, the molecular workhorses that power biofactories! Exciting to see AI accelerating synthetic biology!
“Artios combines backbone and sequence models to achieve state-of-the-art performance in enzyme design.”
Today we are excited to release Artios, a fully open source system for designing functional de novo enzymes.
Enzymes present a new frontier for de novo protein design models due to the challenge of coordinating precise chemistry, rather than binding a target.
Enzymes are life’s biological machines and by designing them we can truly start to engineer biology.
We believe these fundamental capabilities should be built in the open and shared with the community to progress science faster.
🧵 1/6
Ed Harris retweeted
Congratulations on the results @ArdaGoreci. If you are in AI/Bio space, you should follow Ligo Biosciences' work.
Today we are excited to release Artios, a fully open source system for designing functional de novo enzymes.
Enzymes present a new frontier for de novo protein design models due to the challenge of coordinating precise chemistry, rather than binding a target.
Enzymes are life’s biological machines and by designing them we can truly start to engineer biology.
We believe these fundamental capabilities should be built in the open and shared with the community to progress science faster.
🧵 1/6























