@baldur Good to see some down to earth quotes in that article, like how these systems “are designed to make things up. That’s all they do”. I always try to push that ‘hallucinations’ is a bad term because there’s no difference between a ‘hallucination’ and an LLM’s regular generated output - it’s just generating content, the model doesn’t ‘understand’ anything so it can’t understand the concepts of truth or accuracy.

Maybe the output generated for some prompt is accurate, maybe it’s complete nonsense - it’s just chance.

@baldur Wait - OpenAI is *paying* AP for their content?

Certainly sounds like an admission that they should pay for the rights to copyrighted data.

@baldur

I'll be quietly over here getting my work done 20 times faster with language models while everyone can't figure out if they are good or not.

It's a tool. It speeds up certain workflows I have. It also has weaknesses and introduces new attack vectors and error types. We also have not fully explored what we can do with them yet.

But as a 'search' for certain types of exploration of documents it blows traditional search engines out of the water, and you can stack functions to create more complex output.

Most people should probably daily drive them on non critical projects to understand what they can and can't do.

@baldur

“bullshit-making machine makes bullshit” is up there with “dog bites man” and “scorpion stings frog” on the list of the least shocking headlines in history 😛

@baldur exactly this:

(Quote from the article):”””
“This isn’t fixable,” said Emily Bender, a linguistics professor and director of the University of Washington’s Computational Linguistics Laboratory. “It’s inherent in the mismatch between the technology and the proposed use cases.”
“””

It’s like trying to solve the problem that cars don’t provide any nutritional value. The idea of using LLMs to provide detailed factual information is just not tenable. That’s not what they are.

@baldur One of the best ways to predict the next word is to understand the meaning. In some cases, the AI training is going to produce an internal model that actually does match the meaning because that’s going to earn it reinforcement.

But since we can’t really read out what the AI’s reasoning is, we can only train them more and more and can’t tell when they have hit on the correct internal model, and when they are just making a lot of good guesses via approximations. (Ct’d)

@baldur I don't know what else people expected from very advanced autocomplete.

At the end of the day, LLMs are just really, really good autocomplete.

Duck that shirt!

@baldur Of course the marketing firms say "This machine that makes shit up is the best thing ever!" 🙄

I guess it makes sense that the bullshit industry is going to use the bullshit machine.

@baldur

“I probably trust the answers that come out of ChatGPT the least of anybody on Earth,” Altman told the crowd.

Yeah, totally a joke. Right? Right?

*smh* The grift always needs the mark to believe that you can get money for nothing.