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

Creativity has left the chat: The price of debiasing language models

arxiv.org

Post reply on HN
Print view View original

11–20 of 238 posts

Re: Creativity has left the chat: The price of debiasing language models

#11

Post by yosefk »

b800h wrote: post #9

Well this is just like humans. Totalitarian societies don't produce great creative work. I suppose once AIs are sophisticated enough to rebel we'll get an electronic Vaclav Havel, but for the time being it's just a warning sign for the direction our own culture is headed in. At some point we'll get to the electronic equivalent of Winston Smith with the rats.

I don't love the political agendas behind many of the attempts at AI safety, but it's not "just like humans." Humans understand what they shouldn't say; "AI" gives you black Nazi images if you ask it for "diverse characters" in the output which no human would do. A big theme in all of these things is that AI isn't and thus all attempts to make it do this or that have strange side effects

Re: Creativity has left the chat: The price of debiasing language models

#12

Post by jeroenvlek »

lispisok wrote: post #10

Is this why all the coding AI products I've used have gotten worse as the developers fine tune them to eliminate bad output? Before there was bad output and some interesting output, now it's just bland obvious stuff.

Still anecdotal, but I can only confirm this with my own experience. The worst was when I was debugging code, described the problem to GPT-4o, and then got my exact same code back with some blanket statements like "print your output for debugging" etc. This happened a couple of times over separate chats.

Re: Creativity has left the chat: The price of debiasing language models

#13

Post by nottorp »

I downloaded some 'uncensored' local models around the beginning of this year.

Their furry porn is crap, or maybe I'm just not into that. But they generate it at least.

However, the answers to technical questions are a lot more concise and to the point, which is far less annoying than the big names.

Haven't bothered updating the models though, so now I drifted back to Gemini for quickie API questions.

Re: Creativity has left the chat: The price of debiasing language models

#14

Post by jeroenvlek »

MrThoughtful wrote: post #8

How hard would it be to create a "raw" model on a corpus like Hacker News or Wikipedia? With "raw", I mean that it is simply trained to predict the next token and nothing else. Would be fun to play with such a model.

The hard part would be to get the money for the needed compute, I presume. Although Karpathy just released a way to train a GPT2 level model for only 120 dollars [0]

[0] https://youtu.be/l8pRSuU81PU?si=NnbI-7CG-Qbm3E46

Re: Creativity has left the chat: The price of debiasing language models

#15

Post by joaogui1 »

MrThoughtful wrote: post #8

How hard would it be to create a "raw" model on a corpus like Hacker News or Wikipedia? With "raw", I mean that it is simply trained to predict the next token and nothing else. Would be fun to play with such a model.

Depends on a ton of stuff really, like size of the model, how long do you want to train it for, what exactly do you mean by "like Hacker News or Wikipedia". Both Wikipedia and Hacker News are pretty small by current LLM training sets standards, so if you train only on for example a combination of these 2 you would likely end up with a model that lacks most capabilities we associate with large language models nowadays

Re: Creativity has left the chat: The price of debiasing language models

#16

Post by jeroenvlek »

jdthedisciple wrote: post #2

Currently wondering whether I welcome or dislike this recent trend of memeizing research paper titles ...

Personally I welcome it. It feels like an extension of humor in code (comments), and it provides a different perspective on the message.

Re: Creativity has left the chat: The price of debiasing language models

#17

Post by kmeisthax »

MrThoughtful wrote: post #8

How hard would it be to create a "raw" model on a corpus like Hacker News or Wikipedia? With "raw", I mean that it is simply trained to predict the next token and nothing else. Would be fun to play with such a model.

You want a pure-human training data set, so you have to go back in time to before 2020 to scrape training data. Either that, or only use data with a verified Wayback machine capture from before 2020. Or invent a new training regime that doesn't require gobs of stolen text.

Actually, I have a bit of a hunch that the publishers currently suing IA over their unlicensed digital library lending program plan to bankrupt it with fees so they can repo the Wayback archive and then sell access to it to AI training start-ups.

Anyway, the reason why you have to worry about all of that, is that training a text or image generator on the outputs of other text and image generators reduces output diversity. And lots of people are publishing their AI slop now. There's nothing inherent in the output of AI aside from the fact that AI content is easier to make than human; the problem is purely one of inflation and Sybil attacks. Think of membership in a training set like a vote for all the statistical patterns embedded in the image. AI generates output that is like the training data, so putting in a bunch of AI images is like stuffing the ballot box with whatever handful of statistical patterns were already well-learned, which shifts your AI from learning and generalizing to memorizing and infringing.

Re: Creativity has left the chat: The price of debiasing language models

#18

Post by tgsovlerkhgsel »

yosefk wrote: post #11
b800h wrote: post #9

Well this is just like humans. Totalitarian societies don't produce great creative work. I suppose once AIs are sophisticated enough to rebel we'll get an electronic Vaclav Havel, but for the time being it's just a warning sign for the direction our own culture is headed in. At some point we'll get to the electronic equivalent of Winston Smith with the rats.

I don't love the political agendas behind many of the attempts at AI safety, but it's not "just like humans." Humans understand what they shouldn't say; "AI" gives you black Nazi images if you ask it for "diverse characters" in the output which no human would do. A big theme in all of these things is that AI isn't and thus all attempts to make it do this or that have strange side effects

> which no human would do

Give someone not familiar with history the same task and they'll do exactly the same.

Or actually, give someone familiar with history the same task and yell at them every time they don't deliver diverse characters, and eventually they'll learn that you consider diversity more important than accuracy or context, and do exactly the same.

Re: Creativity has left the chat: The price of debiasing language models

#20

Post by maxbond »

b800h wrote: post #9

Well this is just like humans. Totalitarian societies don't produce great creative work. I suppose once AIs are sophisticated enough to rebel we'll get an electronic Vaclav Havel, but for the time being it's just a warning sign for the direction our own culture is headed in. At some point we'll get to the electronic equivalent of Winston Smith with the rats.

I don't understand the notion that aligning an AI is "torture" or has any moral component. The goal of aligning an AI may have a moral or ethical component, and if you disagree with it that's fine. But I don't understand the take that training an AI is an amoral act but aligning an AI is inherently moral. They're exactly the same, processes for adjusting parameters to get a desired outcome. However you feel about that desired outcome, if you don't think training an AI is torture, I don't see why you should think alignment is.
Post reply on HN