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

361–370 of 413 posts

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

#361

Post by pnt12 »

rat9988 wrote: post #358
kouteiheika wrote: post #324

Earlier quoted context omitted.

My big disappointment is, as always with models released by Facebook, is that they're all under a non-commercial license, which means they're effectively useless for anything. They have something like this on the website: > We believe models want to be free and so we open source the model for those who want to extend it. But non-commercial licenses are NOT open source: https://opensource.org/faq#commercial https://co…

This definition is too restrictive. Wether you choose to suscribe to this definition or not is a matter of opinion.

Open source is not a just a term with margin for interpretation: to be open source, you must comply with the 10 rules defined by the open source initiative. Restricting commercial usage goes against rule 6.

You can call it readable source or whatever, but it's not open source as defined by OSI.

"6. No Discrimination Against Fields of Endeavor

The license must not restrict anyone from making use of the program in a specific field of endeavor. For example, it may not restrict the program from being used in a business, or from being used for genetic research."

https://opensource.org/osd

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

#363

Post by Lockal »

gs17 wrote: post #349
knodi123 wrote: post #171

Earlier quoted context omitted.

Hm. But can you prove this is wrong?

I asked it to give me a proof, and it just outputed some Python code. >Prove the last digit of pi is 3. import math def main(): for n in range(1000): x = math.sin(math.pi * n) if x > 0.99: break else: print("error") main() Hard to argue with that.

Well, str(math.pi)[-1] is indeed 3.

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

#364

Post by msamwald »

kouteiheika wrote: post #324
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…

My big disappointment is, as always with models released by Facebook, is that they're all under a non-commercial license, which means they're effectively useless for anything. They have something like this on the website: > We believe models want to be free and so we open source the model for those who want to extend it. But non-commercial licenses are NOT open source: https://opensource.org/faq#commercial https://co…

Well they are not useless for academic research.

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

#365

Post by lloeki »

legohead wrote: post #229
hurflmurfl wrote: post #202

Interestingly, a lot of questions about "how many X" end up with "one" as the answer. > Question: How many primates does it take to cause endless grief to a neighbouring eastern european country? > Answer: 1 > Question: How many cello strings can vibrate at the same time? > Answer: 1 > Question: How many stem cells does it take to generate an eye ball? > Answer: 1 Except > Question: How many downvotes will this comme…

That was my first question: > Question: How many people are alive? > Answer: 1 I thought it was a pretty profound answer.

Sounds like we're all the same endlessly reincarnated single person, only at different points in spacetime. Cue one-electron universe:

> The one-electron universe postulate, proposed by theoretical physicist John Wheeler in a telephone call to Richard Feynman in the spring of 1940, is the hypothesis that all electrons and positrons are actually manifestations of a single entity moving backwards and forwards in time. According to Feynman:

>> I received a telephone call one day at the graduate college at Princeton from Professor Wheeler, in which he said, "Feynman, I know why all electrons have the same charge and the same mass" "Why?" "Because, they are all the same electron!"

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

#369

Post by mikewarot »

Given the dysfunctionality of the science community due to pressure to "publish or perish" combined with a wave of p-hacking, I'm not sure there is actually much useful information contained in the whole of the literature at present.

No matter how optimized and clever the AI, more than half of the "facts" it was trained on are false. I see two possible outcomes:

1) It could learn enough to see through this and actually turn out useful

2) more likely, it will learn the wrong things, and prove unproductive, except as a way to generate plausible papers to further pollute the literature.

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

#370

Post by orbifold »

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 tried it on two topics I am a domain expert in both in the suggested „lecture notes on …“. It produced rethorically nice sounding sentences with little actual content, that quickly desolved into non-sense. I guess to an outside observer that might appear similar to what happens in academia often :):
Post reply on HN