←
Peter Steinbach2,587 posts
Peter Steinbach avatar

Peter Steinbach

@psteinb_

Helmholtz AI Consultants Team Lead for Matter Research @HZDR_Dresden. Opinions are my own.

◷ Joined January 2018▧ 157 Photos and videos
Peter Steinbach avatar
Peter Steinbach@psteinb_·29 Oct 2024

Couldn’t agree more. I hope we @helmholtz_ai can do our share to disseminate SBI to other lands of Physika!

A milestone! 10 years ago I had an epiphany about using machine learning to approximate likelihood ratios, enabling statistical inference when your model is a complex simulator. Since then ~1000 papers on SBI have been published, but this is the 1st from the @ATLASexperiment 🧵

Post attachment
Peter Steinbach avatar
Peter Steinbach@psteinb_·30 Sep 2024

To be even more fair, I’d argue most researchers have a hard time defining what research is (and how it’s done) in the 1st place. Most is done intuitively, people hardly establish methods aside rules of logic. But I agree, a lot of ML lit has flaws. #noprematuregeneralisation

to be fair, much of the published AI literature these days is not research either

Lorena Barba @labarba@fosstodon.org avatar
⟳ Peter Steinbach retweeted
Lorena Barba @labarba@fosstodon.org@LorenaABarba·27 Sep 2024

Very pleased to see this paper now published! Important work, and aligned with my observations. Communicating these concerns often leads to pushback—I know I have been hit by it hard—so kudos to the authors and the journal for publishing it 👏

Our new paper in @NatMachIntell tells a story about how, and why, ML methods for solving PDEs do not work as well as advertised. We find that two reproducibility issues are widespread. As a result, we conclude that ML-for-PDE solving has reached overly optimistic conclusions.

Post attachment
Peter Steinbach avatar
Peter Steinbach@psteinb_·15 Aug 2024

Love the paper and the comment!

Another paper pointing out in details what we've known for a while: LLMs (used via prompting) cannot make sense of situations that substantially differ from the situations found in their training data. Which is to say, LLMs do not possess general intelligence to any meaningful degree. What LLMs can be good for, is to serve as knowledge/routine stores for an actual AGI. They're a memory -- a representation of a data corpus -- and memory is a necessary component of intelligence. But keep in mind that intelligence is not just memory.