maximizing novelty and fun
United States
Joined November 2024
- Tweets7.1K
- Following405
- Followers190
- Likes41.1K
Teaching AI to build scientific models (Anthropic Research, Nov 6, 2026)
A guest post from Anthropic AI for Science grantees Jakob Macke, Daniel Gedon and Stefan Wahl (ELLIS Institute Tübingen / University of Tübingen). Their core claim: AI can now build and fit scientific models fast, but a model that merely reproduces the data gives you an answer, not understanding. What creates understanding is knowing the whole range of models and parameters that fit, and which alternatives fail.
The anchor story is Hodgkin and Huxley's 1951 squid-nerve model. Cambridge's computer was down for six months, so Huxley solved the equations on a hand-cranked calculator, which took over three weeks. To fit the data they had to raise one gating variable to the fourth power without knowing why, and guessed that several charged particles had to move before a gate opened. Decades later (after their 1963 Nobel), the structure of a related potassium channel turned out to have four identical pieces around a central pore. The model, more than 70 years old, still describes neurons from squid to humans.
The authors' point: the fourth-power insight only carried weight because simpler versions demonstrably didn't work. As Richard Hamming put it in 1962, "The purpose of computing is insight, not numbers."
They split modeling into three repeated steps: build (choose components and equation shape), constrain (fit parameters to measurements), interpret (say what it means about the real system). AI is entering all three.
1
34
Constraining. Hand fitting gives one set of numbers and hides all the others that fit just as well. Their analogy: a $120 group dinner bill is consistent with many combinations of orders. Bayesian inference returns the full probability distribution of workable parameters, but computing the likelihood for every combination is intractable for complex models (e.g., thousands of randomly opening ion channels). Simulation-based inference (SBI) sidesteps this: sample random parameters, run the simulator (about 100,000 times for Hodgkin-Huxley, millions for bigger models), then train a neural network that maps an output back to the range of parameters that could have produced it. It only requires that the model can be run forward, even models that are just rules in code. Training takes weeks, but inference on new data then takes seconds; it's fast enough to localize neutron-star collisions in time to point telescopes, and it's used in cosmology, structural biology and geophysics.
The distribution shows which parameters are pinned down, where several explanations remain, and where another experiment would help most.
Building. Historically scientists proposed a handful of candidate models from experience in an essentially infinite space. In the past two years several LLM-agent systems have appeared that take data plus a natural-language description, write model code, fit it, compare to data, revise the equations, and loop. But run an agent twice and you may get two different, equally good models; returning just one hides the rest. So they argue for applying the same distribution idea one level up: the AI proposes many candidate models, each is weighted by how well the data supports it, with models that need lots of flexibility to fit getting less credit, weak ones drop out, and the agent proposes variations of the strong ones.
Features every survivor shares are likely supported by the data; disagreement shows where evidence is thin and what to test next.
Caveat: these systems have mostly been tested on rediscovering known models, with sparse evidence on truly open problems, though newer LLMs keep doing better.
1
24
AI for scientific discovery: How researchers use Claude (Anthropic Economics, Nov 5)
Anthropic's Economics team studied how researchers use Claude. Lead authors are Johannes Hermle, Ben Jones and Lukas Althoff, with several coauthors. Using privacy-preserving classifiers that don't keep the open-ended research content, they labeled 207,049 consumer-plan research sessions from Aug 2 through Sep 30, 2026: 128,914 from Claude (Chat and Cowork) and 78,135 from Claude Code, plus a supplemental sample from Claude Science. Sessions were mapped onto SciNet's taxonomy of 7,262 tasks across 34 fields and 318 subfields.
Which fields use Claude most
• Relative to how many researchers work in each field, formal sciences (computer science, math, statistics) and the social sciences show up more than you'd expect. Lab-heavy physical sciences and medicine show up less.
• On Claude, the largest shares of sessions are computer science at 11.9%, engineering at 8.4% and medicine at 7.7%. Session shares track researcher headcounts fairly closely (correlation 0.79).
• On Claude Code the skew is sharper. Computer science is 4.3% of researchers but 32.7% of sessions (correlation 0.62).
• AI and machine learning alone are 5.0% of Claude sessions and 15.2% of Claude Code sessions, versus 1.2% of researchers. Computational biology, linguistics, neuroscience and robotics also over-index.
2
1
137
What researchers ask Claude to do
• Anthropic sorted work into eight research categories.
• Data analysis is 33.7% of Claude sessions and 48.1% of Claude Code sessions, versus 27.6% of how researchers say they spend their time.
• Writing and communication is 32.0% on Claude versus 16.0% of researcher time.
• Data gathering is underused: 11.5% / 11.9% of sessions versus 28.6% of time, especially physical collection. Digital collection is less underrepresented.
• Ideation and hypothesis generation is only 4.2% on Claude and 2.1% on Claude Code, versus 7.1% of researcher time.
• The single most common task is drafting manuscripts: 18.6% on Claude, 11.1% on Claude Code.
Survey and Claude Science
• A survey of 528 life and health scientists who are heavy AI users matched those patterns. They report the most use for analysis and writing, but still rank ideation first and data gathering second for producing important results.
• On Claude Science, which wires tools and databases into one workflow, biology and biomedical work are more than half of sessions, and ideation rises to about 8%, roughly matching how researchers spend their time. One example: proposing mechanistic models for how proteins interact with each other or with small molecules is 5.6× more common than on regular Claude.
• In biology and biomedicine, Claude Science sessions look longer (about 2× human hours), more doctoral-level (>80% versus about 50–70% elsewhere), and more autonomous (>70% high autonomy versus 23.7% on Claude and 64.1% on Claude Code).
1
1
45
About half a year ago I built my first app, used it for a week or two and then completely remade it so that it works inside grok bot. The app is completely useless with how much AI has advanced, but it was a good problem solving exercise nonetheless.
It’s a simple RSS feed app which can track websites even if they don’t have a feed.
These days Grok Bot sends a detailed summary of the articles I want to read every morning. I’m also going to ask Grok Bot to post that morning briefing on my X account so you can enjoy it too.
The bot will post 1 summary a day, starting from latest -> old. So you might read something that happened a long time ago.
2
1
33
I wish I had never come across this
People not understanding this about Peanuts is a pet peeve of mine. I think people assume it's cute and cheerful like Family Circus or something. But the whole strip is a heart-breaker.
- Charlie Brown is a relentless failure and never vindicated for all his striving
- Linus's faith is never rewarded
- Lucy's love is unrequited and she becomes worse for it
- Peppermint Patty's love for Charlie Brown is never returned
and on and on
Snoopy escapes from this through his imagination. He's totally in his own world and lives by his own rules. The cruelty of life can't get to him.
94
David’s podcasts always make me reflect so here goes
When I was young I used to think it’s impossible to work like this but now I realize it’s not. The only thing is you have to be working towards things that you truly believe are important. I used to spend 15 hours a day gaming when I was a kid. How is that not different from the effort Oda is putting. I spent 15 hours a day studying finance in college because I wanted to.
The idea I relate with the most is creating for an audience of 1. Yourself. This is such an insanely double edged sword but it truly brings out something primal out of you. When I played video games I always thought I could have managed the games better than the devs. They always took decisions I hated. All the games I’m talking about have completely abandoned their existing fanbases to attract a new audience. Perhaps because of survival. Idk. Half are dead. Half are barely alive. I know feedback is important, but I still cant help but cringe when I see people asking for feedback.
Now with AI it’s affordable to create for an audience of 1. This is the greatest opportunity for you to push your taste onto other people.
A few surprising things I learned from reading about Eiichrio Oda, the creator of One Piece:
1. He slept 3 hours a night for 15 years.
2. He created the most successful manga of all time with more than 600 million copies in circulation.
3. He said he would have missed his parent’s funeral because it conflicted with his work.
4. He is extremely private. He does not allow his face to be photographed and when he appears on video he has a giant fish avatar placed on his head.
5. When he hires a new editor the first thing he says to them is you need to be willing to die for One Piece.
6. His editor said he might someday encounter a writer more talented than Oda but he would never meet anyone who worked harder.
7. He doesn’t want ideas, he wants criticism.
8. He would hand draw entire Disney animated movies to learn how to give his characters emotion.
9. He said he only saw his family once a week because family time was too relaxing.
10. He holds the Guinness World Record for the most copies published for the same comic book series by a single author.
11. He is often described as a maniac due to his unbelievable work ethic and inhuman daily habits.
12. He believes in having the least number of people possible touching the work.
13. He says he gets his best ideas by not sleeping or eating. He says humans can only come up with truly new ideas when they reach their limit.
14. He started working on One Piece when he was 22. He is still working on it at 51.
15. He once spent 12 consecutive hours on the telephone discussing a single story with his editor.
16. He has an extreme tolerance for repetition.
17. Imagination is the organizing force of his entire life.
18. He's a perfectionist who has repeatedly admitted that he cannot make himself work in a more efficient way, even when he desperately wants to.
19. He never finishes before a deadline. He equates using every available minute with giving his maximum effort.
20. He believes it’s impossible to create something great unless you do everything yourself, attend to every detail, and test it exhaustively.
21. He believes the more experience he has the harder it is to create.
22. He was 4 when he decided to be a manga artist. He said he wanted to avoid ever having to get a real job.
23. He creates his work for an audience of exactly one person: himself at 15 years old.
25. He has dedicated his entire adult life to writing a story about freedom and adventure, while accepting an extraordinarily restrictive life in order to create it.
26. To this day he does not feel successful.
--
Eiichiro Oda is a relentlessly obsessive creative genius who has devoted nearly every waking hour of his adult life to building one of the most extraordinary fictional universes ever imagined. You don't want to miss this episode:
4
122
What’s also not talked about enough is how Cursor went from having the best Cloud/Remote experience to having the worst. And not much has changed, except that the competitor’s models are so far ahead that it feels like you’re sifting through garbage when you use the Cursor application.
If you can’t compete on models, you’re nothing. Also no wonder Anthropic has partnered so closely with SpaceX. They are literally not a threat.
Cursor cloud is just completely useless. Even with Claude writing a plan telling it which exact lines of code to replace and what to replace them with, it will run for 30 minutes and do a terrible job. I meed to figure out how I can get my money back from Cursor.
1
29
Uhh is apple no longer worth a trillion dollars what’s going on
NEW: Apple is cutting employee AI spending budgets even as it pushes workers to use the technology more. Engineers have seen daily limits cut from $300 to $200, with further reductions to $100 planned. Other teams could see limits fall to $25. bloomberg.com/news/articles/…
38
Used GPT today after a loooong time to come up with brand names and the free version did a much better job with worse prompts than Opus 5.5. Curious…
1
26
Getting the Annual X subscription was honestly the worst decision of my life
I feel handicapped when my claude usage runs out
21
Claude Opus 5.5 is so good that I am changing the way I work so I never have to leave the Claude ecosystem
11
