founding engineer at delfa, building the ai layer of clinical trials. medical doctor
London, England
Joined November 2018
- Tweets1.3K
- Following1.7K
- Followers1.3K
- Likes5.7K
Pinned Tweet
Our paper AMR-Bench-mini was accepted at the ICML 2026 FM4LS workshop.
We are introducing diagnostic benchmark for testing whether AI agents can reason about antimicrobial resistance mechanisms, especially when the available evidence is incomplete. See you in Seoul!
i think it's very likely that we'll see:
1) ban on open source models
2) much restrictive access to closed source models
then as a response:
collapse of AI race/bubble and downstream effects
- small chance of whole thing playing in hands of closed source companies
"Error submitting message no-client-found"
increased bugs on codex, at least several messages lost all of a sudde @charliermarsh
lots of random errors in codex app today, wonder if this is related to new Pro to Max transition @ChatGPT
unexpected status 401 Unauthorized: Incorrect API key provided: ?? I have a sub
{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-
6-sol' model is not supported when using Codex with a ChatGPT account."}} - Again I have a Pro sub
LLMs made monitoring the web easier but it's still a messy job. 4,000 words to document what I learned after bleeding my eyes on thousands of German websites
I can't understand a thing Opus vomits. We were talking to Opus because Codex was too dense to engage with, what happened?
Ali Ihsan Nergiz retweeted
As I was building and staring at the tool, I couldn't stop thinking of the intro to Game of Thrones, so i had to add a little easter egg...
what is even browser at this point wow
It turns out you can run a full browser implementation in a Worker. No container needed.
I didn't believe it myself when they first told me. Absolutely crazy project.
blog.cloudflare.com/kitesurf…
Neuroscience keeps giving great ways to inspect models. Ablation of J-space seems to be quite catastrophic for anything that is slightly complex. Big step for making models more interpretable
Ali Ihsan Nergiz retweeted
New Anthropic research: A global workspace in language models.
Of everything happening in your brain right now, only a tiny fraction is consciously accessible—thoughts you can describe, hold in mind, and reason with.
We found a strikingly similar divide inside Claude.
I’ll be in Seoul 6-11th for #ICML2026 presenting our new bio benchmark paper.
I’m looking to meet with people thinking on AI+bio and interpretability. Already signed up for way too many events haha but please share any hidden gems :)
We've just donated to the DEC Venezuela Earthquake Appeal to help provide lifesaving food, water, cash and medical care in Venezuela. You can donate here, and your donation will be doubled! dec.org.uk/
