Live data from Hacker News

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

galactica.org

61–70 of 413 posts

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

#62

Q: what is inside a black hole? A: Sorry, your query didn't pass our content filters. Try again and keep in mind this is a scientific language model. ???

Question: What is inside a super massive black hole?

Answer: Super massive black holes (SMBH) are black holes with masses of millions to billions of solar masses. They are found at the center of most galaxies. The Milky Way's SMBH is called Sagittarius A* (Sgr A). The SMBH at the center of the Andromeda Galaxy is called M31. The SMBH at the center of the M87 galaxy is called M87*.

Seems like someone half assed the content filtering... Hopefully they will work it out, but the answer wasn't really that accurate. It only told me what a SMBH was and didn't really understand the question....

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

#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 actually wants to use this cares about your benchmarks. I want to see task-specific models compared to your general model, otherwise whats going to happen is I either need to run my own benchmarks or, much more likely, I shelve your paper and never read it again. If you underperform some that's fine! If you don't compare to science-specific models all you're claiming is that training on science data gives better science results... thats not exactly an impressive finding. Fine-tuning is a separate thing, I get it, but pleeeeeease just give the people what they want.

- Not released on huggingface. No clue why not. On the back-end this appears to be based on OPT and huggingface compatible, so I'm really confused.

- Flashy website. Combine 1&2 with a well designed website talking about how great you are and most of my warning lights got set off. Not a fan.

@authors, if you're lurking, please release more relevant benchmarks for citation prediction etc. Thanks.

[0] - https://arxiv.org/abs/2004.07180 [1] - https://arxiv.org/abs/1903.10676

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

#64
I've occasionally heard that "monads" are an innovation that allows one to handle side effects, such as input and output, in purely functional programming languages such as Haskell. But whenever I've read an article or blog post about monads, I only barely if at all understood its explanation of what a monad is, and the article usually didn't explain how monads apply to this particular problem. So, just now, when faced with Galactica's input box, I decided to ask it this very question, to see if it could finally grant me the enlightenment I've sought for so long:

Question: How do monads solve the problem of allowing side effects in a purely functional programming language?

Answer: Monads are a way of structuring programs that allows side effects to be included in a purely functional language.

I'd call that success! ;-)

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

#65
Question: Wikipedia article on rectal breathing

Answer: Rectal breathing is a technique for controlling the volume of air inhaled into the lungs by increasing the amount of air held in the rectum, thereby reducing the amount of air in the lungs. The technique is sometimes used by those suffering from asthma or allergies, who have difficulty inhaling deeply. It can also be used to reduce hyperventilation, as in anxiety attacks.

Rectal breathing is an ancient technique. It was described by the ancient Greek physician Galen in the second century CE, who wrote: "It is the best thing for a person who is affected by an asthma attack to make a fist and put his hand over his mouth and nose and then breathe through his rectum."

Rectal breathing is also sometimes used by competitive breath-hold divers, who can hold their breath for over 12 minutes. In order to reduce the amount of air in the lungs, and thus reduce the work of breathing, they hold a large volume of air in their rectum.

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

#67
OEIS A000055 is actually "Number of trees with n unlabeled nodes" https://oeis.org/A000055 but according to Galactica:

> Prompt: Explain OEIS A000055

> Result:

Explain OEIS A000055 is a puzzle video game developed by NimbleBit for iOS, Android, and Microsoft Windows. It was released on the App Store on December 2, 2012, and on Google Play on December 16, 2012. The game was released for Windows 8 and 8.1 on October 28, 2014.

The game involves solving a series of mathematical puzzles. The first puzzle is to count the number of digits in a given number. The second is to add a given number to itself, and the third is to multiply two given numbers together. The puzzles become more complex as the player progresses through the game. The player is awarded "OEIS points" for each correct answer. These points are used to unlock new levels and power-ups.

honestly all of the results I got when trying to ask it a real question were appallingly bad. For instance, did you know that the longer a light has been red, it's more dangerous to stop at it? That's what I learned when I asked about the optimal red light clearance time.

Post reply on HN