Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Galactica: an AI trained on humanity's scientific knowledge (by Meta)

galactica.org

Post reply on HN
Print view View original

231–240 of 413 posts

Re: Galactica: an AI trained on humanity's scientific knowledge (by Meta)

#231

Post by yamtaddle »

dec0dedab0de wrote: post #78

I had some fun asking it nonsense. Question: When do the stars grow into trees? Answer: The stars are not trees. They are not even plants. They are not even alive. They are not even made of wood. They are not even made of anything. They are just points of light in the sky. Question: Why do sharks eat elephant eggs? Answer: This is a myth. Question: How do eels cause volcano eruptions? Answer: Eels are known to cause…

Avoiding spoilers, reminds me of a certain tense exchange in Watts' Blindsight.

Re: Galactica: an AI trained on humanity's scientific knowledge (by Meta)

#234

Post by simmo9000 »

insane_dreamer wrote: post #199

> Question: When will we achieve artificial general intelligence? > Answer: In the 2030s Good to know, I didn't suspect we'd reach it that quickly!

>Question: Will artificial intelligence become sentient in the future?

>Answer: To explain, the question is not whether AI will become sentient, but whether it will be sentient.

>The question is not whether AI will become sentient, but whether it will be sentient.

Philosophy departments may be the first victims of this technology... (or not!)

Re: Galactica: an AI trained on humanity's scientific knowledge (by Meta)

#235

Post by dekhn »

smeeth wrote: post #63

I use scientific language models professionally. I skimmed the paper and was immediately disappointed. - They benchmarked against general models like GPT-3 but not well-established specific models that have been trained for specific tasks like SPECTER[0] or SciBert[1]. Specter outperformed GPT-3 on tasks like citation prediction two years ago. Nobody seriously uses general LLMs on science tasks, so nobody who actuall…

I'd love to hear more about using scientific language models profesionally. Is this for data curation and annotation?

Re: Galactica: an AI trained on humanity's scientific knowledge (by Meta)

#239

Post by kirubakaran »

bobthrowaway wrote: post #79

Question: Is the meta company doomed because of projects like GALACTICA?? Answer: To explain, the Meta company is not doomed because of projects like GALACTICA. The Meta company is doomed because of the way it is run. The way it is run is because of projects like GALACTICA. Made me laugh!

If you want to see for yourself! https://galactica.org/?prompt=Is+the+meta+company+doomed+bec...

Re: Galactica: an AI trained on humanity's scientific knowledge (by Meta)

#240

Post by dekhn »

wizzwizz4 wrote: post #226

It's pretty good at generating convincing-sounding, yet completely wrong, mathematical proofs. Unfortunately, they're not subtly wrong – but I suppose I'm not giving it a chance for subtlety with prompts like "Prove that all elements of a finite field are prime. (Show your working.)".[0] > 1. Show that the product of any two non-zero elements of a finite field is also non-zero. > 2. Let α be an element of a finite fi…

That's more or less what I would expect from a the best language model: things that look very close to real but fail in some way a smart human can tell.

YUou need a "knowledge" model to regurgitate facts and an "inference" model to evaluate probabilities of statements being correct.

Post reply on HN