Scientists begin building AI for scientific discovery using tech behind ChatGPT | Hacker News Reader

Scientists begin building AI for scientific discovery using tech behind ChatGPT

techxplore.com

40 pointsgardenfelder15 comments

I am trying to think of a reason for why there are many articles announcing the start of a grand initiative, but few that would announce one coming to fruition. Can't quite put my finger on it.

They need content, but what do you mean they wouldn’t announce one coming to fruition? I bet they would if it was a slow news day, or just depending on what cooler things are available.

This is business 101. Sell the idea before you have the product. Then try to create the product with the funds raised (or not).

As your sass implies, it could be this may all be hype. But also if it wasn't it would take a while for exciting results to arrive.

I’m doing this for computational chemistry

https://atomictessellator.com

Github https://github.com/PolymathicAI

I’m fairly skeptical about AI hype, but there are probably plenty of applications of the tech for use in scientific discovery. Just not the “Albert Einstein bot” or whatever that gives you fully formed scientific theories.

More like tools to search large spaces for theories/theorems that seem likely that can then be independently verified. As I understand it we’ve been using AI in space exploration/astronomy for a long time to search data generated from telescopes I can imagine there are analogous uses for chatGPT style solutions.

"An international team of scientists, including from the University of Cambridge, have launched a new research collaboration that will leverage the same technology behind ChatGPT to build an AI-powered tool for scientific discovery."

the BLOOM team is drinking their wine while you toil, Cambridge?

https://en.wikipedia.org/wiki/BLOOM_(language_model)

They can't probably join BLOOM collaboration, while BLOOM has already released a model and probably will not receive anymore funding. BLOOM model is the only model that spooked me by answering "I do not know. I do not know. I do not know..." on some of my questions.

Are LLMs capable of any form of innovation or discovery that is not a simple mix of existing concepts?

(It could be argued that all innovation or discovery is a simple mix of existing concepts, but I don't believe this to be true).

Whether they are or they aren't, we can get very far with discovery that is a simple mix of existing concepts

Absolutely! And good science is sometimes just about organizing existing information too, which I believe LLMs can excel at.

The word "simple" is really loaded here. Is the universe a simple mix of existing physics? Is a 70B parameter matrix of numbers containing the relative relationships between a significant chunk of published human knowledge actually that simple?

Sure, but I have yet to see LLMs actually innovate and not just pretend to be innovative.

Reply on news.ycombinator.com