Alex Rubinsteyn

Scientific computing in the age of agentic AI: openai.com/index/scient... (we have a mhcflurry vignette in this)

openai.com

really interesting thread on X about the latest batch of LLM proofs for open math problems, these centered on theoretical computer science / coding theory / quantum computing / &c x.com/henryquantum...

Henry Yuen (@henryquantum) on X x.com

Common in the 1960s to conflate computing with cognition and underestimate how long it would take to get from Perceptron to something like modern AI by many decades. But the worries about impact on society and economy same as today (Reading “Cybernation: The Silent Conquest”)

babel.hathitrust.org

So many published bio ML models score within like 10% of a naive linear baseline

I have an important confession that I need to get off my chest, of great relevance to the scientific community and society in general. I have, and this is hard to admit publicly, been using AI.

Listen, I asked Astra to scrape a website for genomics test cases & it honed in on the most interesting mutations, downloaded many GB of seq data, pulled out subsets around loci of interest & turned them into well constructed tests Your theory of AI has to at least acknowledge that it’s magic

AI companies turning to biomedical research is awesome for humanity but bearish for AI companies

Tried a code review from GPT 5.6 sol after using Opus 4.8 and GPT 5.5 for a while and it was a qualitative step up wrt insight, high level understanding of intent, and anticipating future failure modes Maybe Fable would be as good but still can't try it with anything bio...

Astra finding bugs in everything Not as severe as when I switched to agentic coding and Claude Code found more severe bugs in all my hand-written tools, but still surprised by the volume

Computational Medicine program at UNC is doing a faculty search and I really want more translational / AI-centric labs to join the floor. Anyone on the market who wants to start clinical trials on short timelines and uses AI in the clinical loop?

I guess the last hiding place for “gen AI can’t do -real- math” is interesting new definitions. Is there an interesting AI proof that isn’t either pulling from another (maybe distant) field or enumerating in some way?

Navier Stokes is just gossip at this point but after using Astra a little it’s imaginable that many of our most famous math problems fall to AI proofs in the next 2-3 years. What then? Do we move on to newer more difficult formal systems?

Tried a code review from GPT 5.6 sol after using Opus 4.8 and GPT 5.5 for a while and it was a qualitative step up wrt insight, high level understanding of intent, and anticipating future failure modes Maybe Fable would be as good but still can't try it with anything bio...