New study shows LLMs outperform neuroscience experts at predicting experimental results in advance of experiments (86% vs 63% accuracy). They use a fine-tuned Mistral 7B but other models worked too. Suggests LLMs can integrate scientific knowledge at scale to support research. - ThreadSky

emollick.bsky.social • 91 days ago

New study shows LLMs outperform neuroscience experts at predicting experimental results in advance of experiments (86% vs 63% accuracy).

They use a fine-tuned Mistral 7B but other models worked too. Suggests LLMs can integrate scientific knowledge at scale to support research.

1 / 3

Comments

janbam.bsky.social•91 days ago

biased

caracter.bsky.social•91 days ago

Just my two cents.
So basically you are saying LLMs predict the study results before it happened?

How is that significant? Easn't the point of LLMs to predict the most likely token? So I would expect them to very good at anything related to prediction.

themajor.bsky.social•91 days ago

... Hinton makes this point alot ... still getting my head around it

johnrecords.bsky.social•90 days ago

Impressive that a 7B model could do this.

slavav.bsky.social•91 days ago

Insane that 7b model achieves that

yshemesh.bsky.social•91 days ago

Given LLMs model multiple perspectives in the training data, they may express a sort of wisdom-of-crowds (higher accuracy than any one individual). But this may also mean worse predictions on surprising/exceptional results. If true, this would strongly motivate against LLMs to guide research agenda

yshemesh.bsky.social•90 days ago

For example: https://open.substack.com/pub/understandingai/p/why-the-deep-learning-boom-caught?utm_source=share&utm_medium=android&r=2rbsd @binarybits.bsky.social

dmewes.com•91 days ago

One thing I'd expect LLMs to be better at than humans is being calibrated correctly to the statistical distribution of the two possible outcomes, as they will resemble max likelihood prediction. Maybe human scientists could improve their accuracy if they first went through calibration training?

dmewes.com•91 days ago

One way to validate this would be to ask humans to provide a 0-1 likelihood for one of the binary experiment outcomes, instead of asking for a binary choice Then determine the max-likelihood threshold based on a calibration ("training") set for each individual human rater to build a human classifier

mh123.bsky.social•91 days ago

To be fair: No experiments were conducted in this study. LLM discriminated between published abstract or an altered version of it. Training data was reasonably shown not to be contaminated by said abstracts.

gnperdue.bsky.social•91 days ago

That is quite different from what is implied by the abstract of this paper. I haven’t read this paper, but how did they audit the training data to be sure it didn’t contain the papers from this study?

emollick.bsky.social•91 days ago

gnperdue.bsky.social•91 days ago

Thanks.

emollick.bsky.social•91 days ago

https://www.nature.com/articles/s41562-024-02046-9.pdf

grahamerwin.bsky.social•90 days ago

Thanks Ethan! The two things that caught me were 1) fine-tuning only gets you an additional 3% accuracy and 2) "instruct" models perform worse than base models! The authors go on to say "...aligning LLMs to engage in natural language conversations hinders their scientific inference abilities."

docvalerie.bsky.social•90 days ago

Thanks for this.

IMO though, this is not surprising.
Scientists have become increasingly overspecialized in the past 30 years, while LLM's through their training properties become rather, generalists.

Take a sample of more generalist scientists who practice across interdisciplinary fields & try.

aschoi.bsky.social•91 days ago

I mean, isn't it possible that LLMs could just be better at reading scientific abstracts? Not necessarily better at predicting neuroscience results?

pheasant.bsky.social•90 days ago

Better at predicting what an abstract should look like. The scientists are looking for novelty, the LLM is looking for patterns

yo-cuddles.bsky.social•91 days ago

This is probably a function of LLM's superhuman memory. Not strict memorization, which seems controlled for, but I imagine many of these questions were pretty predictable to specialists. LLM's would prob score do better on a test that combined tests for doctors and lawyers than almost any human

Comments

Posting Rules

Reply