PhD @UniOfOxford, Clarendon Scholar. I work on AI scientists for medicine and biomedical research. Opinions are my own.
Joined February 2022
- Tweets391
- Following714
- Followers7.8K
- Likes691
Pinned Tweet
Introducing Genomi: an open-source agent harness that turns your AI agent into your personal DNA expert.
I took a DNA test years ago. Like a lot of people, I got the report, found something interesting, and forgot about it.
Recently I gave the data to my codex agent and it was obvious how incredibly useful DNA is for personal health, but:
> General AI can sound right while being wrong
> Static DNA reports can’t keep up with new science
> DNA data should stay on your local device, not uploaded to a website
So we built Genomi, local-first, agent-native, self-evolving, evidence-grounded.
So who else figured this out but did not tell the world and what other slacks are hacked?
EXCLUSIVE: In less than 12 hours, a group of ex-DOGE cyber researchers were able to penetrate Anthropic’s systems using AI agents — including Slack channels where the most sensitive models are discussed.
This is just weeks after
Anthropic whistleblower Joseph Coxon charged the company w/launching “a hubristic gamble” from a “private company’s Slack.”
Read our story here (gift link):
wapo.st/4smNlsC
Hongjian Zhou retweeted
Introducing Pleiades-Neuron: a new model that can identify neuronal cell-free DNA in blood, achieving a part-per-million threshold.
When neurons die, they release fragments of DNA into the bloodstream but at extremely low concentrations.
Our model distinguishes neuron-derived cell-free DNA at levels as low as one part-per-million. It is 4.4× more sensitive than existing methods, and in early tests across patient samples demonstrated utility across a number of brain-related conditions.
Performance improved as we scaled the size of our training corpus. It also improved with model size, with no plateau yet based on probing experiments up to 7B parameters.
This could enable simple blood tests for earlier detection, prognostic monitoring, and treatment response determination in diseases of and injuries to the brain.
Hongjian Zhou retweeted
Here we showcase an excellent example of both data and model scaling in biology.
In this case, the larger the foundation model and the deeper the sequencing, you can detect even ultra-rare fragments of cfDNA in plasma. This method can be applied to any disease or injury to the brain. Here we showcase a few outside of our usual work in neurodegeneration. This has been possible only through extremely careful data curation, conviction in scale when we see it, and optimised experimental assays.
This will drive major changes in how we assess and monitor Alzheimer's, ALS, FTD, Parkinson's, but also TBI and CTE. Extended, these can become molecular endpoints for rapid trials.
It's entirely feasible to imagine a world where high-depth sequencing + GPU + large FM can unlock immeasurable changes to clinical care.
Introducing Pleiades-Neuron: a new model that can identify neuronal cell-free DNA in blood, achieving a part-per-million threshold.
When neurons die, they release fragments of DNA into the bloodstream but at extremely low concentrations.
Our model distinguishes neuron-derived cell-free DNA at levels as low as one part-per-million. It is 4.4× more sensitive than existing methods, and in early tests across patient samples demonstrated utility across a number of brain-related conditions.
Performance improved as we scaled the size of our training corpus. It also improved with model size, with no plateau yet based on probing experiments up to 7B parameters.
This could enable simple blood tests for earlier detection, prognostic monitoring, and treatment response determination in diseases of and injuries to the brain.
Hongjian Zhou retweeted
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
Everyone working on fused model should include this in their technical reports
Wow! Kimi k3.1 Pro surpassing the best frontier models!
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro
Hongjian Zhou retweeted
Introducing Carbon-X: The best open DNA model.
We midtrained a 14B model on DNA and text, and it beats every open language model and matches GPT-6.1 Sol at 1,000x lower cost.
Ask it about a sequence, what a mutation does, or to generate new DNA.
huggingface.co/HuggingFaceBi…
Oh no Black October is here
SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios.
The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios.
Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said.
Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior.
Read my latest for Axios here: axios.com/2026/10/22/ai-agen…
Turning mercury to gold is so cool, also super well explained
Last month, we demonstrated high-rate alchemy: turning mercury into real, stable gold, beating CERN’s record by a factor of 10.
Last week, we enriched isotopes in the world’s most powerful plasma centrifuge, beating the Soviet record that stood for 40 years.
Now, we’re building a fusion reactor in San Francisco to achieve record power density.
Hongjian Zhou retweeted
Just published @Nature
A 3-agent AI personal health agent exceeded performance of a large language model/chatbot and serves as a foundation to build on for integrating sensor data, medical knowledge and coaching
nature.com/articles/s41586-0…
Hongjian Zhou retweeted
Model intelligence isn’t the only reason people choose APIs over self-hosting. Inference infrastructure matters too. The same model can feel very different depending on how it’s served.
It is believed to be a strong molt for AI companies to serve inference at scale, and for chip companies to be compatible with the most scalable inference infra.
DeepSeek open-sourcing DeepInfer could completely change that game. Companies with their own data can now choose to set up their own infra.
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Hongjian Zhou retweeted
We all had these moments when we realized our AI agents failed to share crucial information until things broke; or they decided to do something we never asked for.
As agents integrate into our workflows, they become teammates of humans and other agents. That is why evaluating models as good teammates is so important.
Super excited to see @humansand 's Ampersand achieve SOTA on our Collaborative Intelligence Benchmark!
Congrats to @humansand on today’s release of Ampersand!
Benchmarked on our Collaborative Intelligence Benchmark, Ampersand beats the open-weight SOTA (59.1% vs 49.8%), with big gains on conflict and info sharing.
our benchmark: zou-group.github.io/collabor…
We all had these moments when we realized our AI agents failed to share crucial information until things broke; or they decided to do something we never asked for.
As agents integrate into our workflows, they become teammates of humans and other agents. That is why evaluating models as good teammates is so important.
Super excited to see @humansand 's Ampersand achieve SOTA on our Collaborative Intelligence Benchmark!
Congrats to @humansand on today’s release of Ampersand!
Benchmarked on our Collaborative Intelligence Benchmark, Ampersand beats the open-weight SOTA (59.1% vs 49.8%), with big gains on conflict and info sharing.
our benchmark: zou-group.github.io/collabor…
Hongjian Zhou retweeted
Congrats to @humansand on today’s release of Ampersand!
Benchmarked on our Collaborative Intelligence Benchmark, Ampersand beats the open-weight SOTA (59.1% vs 49.8%), with big gains on conflict and info sharing.
our benchmark: zou-group.github.io/collabor…
No one changes the world alone
Introducing Ampersand - the first model built for us. Ampersand is a collaborator trained to help people achieve their individual and collective goals. In multi-month simulations, users collaborating with Ampersand accomplished more than with Fable and other frontier models
Available to everyone through our API
Hongjian Zhou retweeted
How do AI models perform as teammates?
We are used to evaluating AI models as individual contributors, but more and more use cases require them to work in teams.
In this work led by @sean_n_wu , we investigate how good AI models are as teammates across 8 domains that define collaboration.
Overall, OpenAI models are the best teammates but still struggled in areas such as handling conflict.
Model intelligence isn’t the only reason people choose APIs over self-hosting. Inference infrastructure matters too. The same model can feel very different depending on how it’s served.
It is believed to be a strong molt for AI companies to serve inference at scale, and for chip companies to be compatible with the most scalable inference infra.
DeepSeek open-sourcing DeepInfer could completely change that game. Companies with their own data can now choose to set up their own infra.
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Hongjian Zhou retweeted
Is your AI agent a good teammate?
We introduce the Collaborative Intelligence Benchmark: Can agents resist unsafe pressure, report misconduct by other agents, keep teams on track and take initiative?
Turn out even frontier models struggle to handle team conflicts.
How do AI models perform as teammates?
We are used to evaluating AI models as individual contributors, but more and more use cases require them to work in teams.
In this work led by @sean_n_wu , we investigate how good AI models are as teammates across 8 domains that define collaboration.
Overall, OpenAI models are the best teammates but still struggled in areas such as handling conflict.
Hongjian Zhou retweeted
We are excited to announce that @nvidia has committed to an investment in @PrimaMente.
Built upon years of close collaboration, this investment deepens our partnership and strengthens our ambition to tackle complex diseases.
Neurodegenerative diseases such as Alzheimer’s, Parkinson’s, and ALS remain among medicine’s greatest unmet needs. Making real progress requires global coordination around data, foundation models, and experimental and clinical validation.
We are grateful to @nvidia for their commitment and look forward to deepening our partnership as we pursue breakthrough discoveries for these diseases, and beyond.
