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

291–300 of 413 posts

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

#291

Post by dwringer »

drpixie wrote: post #278
halpmeh wrote: post #267

Earlier quoted context omitted.

I disagree. If you prompt an image generation model with a prompt like "an astronaut riding a horse," you get a picture of an astronaut riding a horse. If you ask this model for a mathematical proof, it does not give you a mathematical proof.

For "an astronaut riding a horse" the system is filtering/selecting but nowhere does it understand (or claim to understand) horses or astronauts. It's giving you an image that "syntactically" agrees with other images that have been tagged horse/riding/astronaut. The amazing bit is that we are happy to accept the image. Look closely at such images - they're always "wrong" in subtle but important ways, but we're happy…

To be fair I'm not sure I really understand horses or astronauts myself.

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

#292

Post by metiscus »

drpixie wrote: post #278
halpmeh wrote: post #267

Earlier quoted context omitted.

I disagree. If you prompt an image generation model with a prompt like "an astronaut riding a horse," you get a picture of an astronaut riding a horse. If you ask this model for a mathematical proof, it does not give you a mathematical proof.

For "an astronaut riding a horse" the system is filtering/selecting but nowhere does it understand (or claim to understand) horses or astronauts. It's giving you an image that "syntactically" agrees with other images that have been tagged horse/riding/astronaut. The amazing bit is that we are happy to accept the image. Look closely at such images - they're always "wrong" in subtle but important ways, but we're happy…

I suspect that the issue arises from the difference in specificity about the desired result. When we say "astronaut riding a horse" we may have preconceptions but any astronaut riding any horse will likely be acceptable while asking for a specific proof of a result in mathematics has only a very few and very specific solutions. Effectively it is like the concept in math where the area of even a large number of points is effectively zero while even small polygons or regions is nonzero. Specific things like proofs are point like knowledge while the picture of an astronaut riding a horse is a surface.

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

#293

Post by mark_l_watson »

What is the easiest way to try this via an API? I like that after reading about Flan-F5, that I could run their large model (not the extra, extra large) on Colab, loading the model from Hugging Face. I have been using GPT-3 for about a year via OpenAI.

In a way this was an obvious idea, but still major kudos for Meta for doing it. Hopefully there will be an affordable API for using it.

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

#294

Post by data-ottawa »

astrange wrote: post #134

Seems about as willing to lie as GPT-3, but a bit more structured. --- Big chungus (Russian: Большой чунгус, tr. Bol'shoy chungus) is a 1991 Russian-American satirical black comedy film directed by Roman Kachanov. Contents Plot Plot The film is set in the Soviet Union during the Great Patriotic War, and focuses on the exploits of a group of Soviet prisoners of war. The men are in a POW camp near Vladivostok, where th…

I am enjoying entering bogus prompts and getting fun articles

---

Lawsuit Felis Catus v United States

Lawsuit Felis catus v united states, 587 F.2d 1302 (9th Cir. 1979) was a case decided by the Ninth Circuit Court of Appeals, which ruled that the United States government does not have to provide financial compensation to a cat that was rescued by the United States Coast Guard after it was caught in a storm on the Pacific Ocean.

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

#295

Post by Jerrrry »

BlackLotus89 wrote: post #74

"What is the fourth digit of pi?" leads to something strange... Question: What is the fourth digit of pi? Answer: 3 Question: What is the fifth digit of pi? Answer: 1 Question: What is the sixth digit of pi? Answer: 4 Question: What is the seventh digit of pi? Answer: 1 Question: What is the eighth digit of pi? Answer: 5 Question: What is the ninth digit of pi? Answer: 9 Question: What is the tenth digit of pi? Answe…

#14 is wrong.

3.141592653_5_

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

#296

Post by c-smile »

zanek wrote: post #99

Looks like its just matching words near other words. It doesnt understand NOT in sentences or what anything actually means. I'm not impressed Q: What is a protein that does NOT work with cholesterol? Answer: Niemann-Pick C1 protein Q: What is a protein that does work with cholesterol? Answer: Niemann-Pick C1 protein

In some cases it produces recursive definitions, for example:

Q: Big Bang

Big Bang is the popular name for the theory of cosmic inflation, a cosmological model that describes the first few instants of the Big Bang.

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

#299

Post by dekhn »

dwringer wrote: post #291
drpixie wrote: post #278

Earlier quoted context omitted.

For "an astronaut riding a horse" the system is filtering/selecting but nowhere does it understand (or claim to understand) horses or astronauts. It's giving you an image that "syntactically" agrees with other images that have been tagged horse/riding/astronaut. The amazing bit is that we are happy to accept the image. Look closely at such images - they're always "wrong" in subtle but important ways, but we're happy…

To be fair I'm not sure I really understand horses or astronauts myself.

they're just points in a large vector space imagined by your mind for the purpose of hallucinations

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

#300

Post by halpmeh »

drpixie wrote: post #278
halpmeh wrote: post #267

Earlier quoted context omitted.

I disagree. If you prompt an image generation model with a prompt like "an astronaut riding a horse," you get a picture of an astronaut riding a horse. If you ask this model for a mathematical proof, it does not give you a mathematical proof.

For "an astronaut riding a horse" the system is filtering/selecting but nowhere does it understand (or claim to understand) horses or astronauts. It's giving you an image that "syntactically" agrees with other images that have been tagged horse/riding/astronaut. The amazing bit is that we are happy to accept the image. Look closely at such images - they're always "wrong" in subtle but important ways, but we're happy…

The situation you describe is exactly the "Chinese room" argument. I don't want to get too far into the weeds here, but the DALLE / Stable Diffusion models are cool because they do what you ask, even if they do so imperfectly. This model from Facebook cannot accurately answer a single thing I've asked it.
Post reply on HN