@Gargron yep, a fake news machine. Anybody thought about feeding the chat bot’s stuff to an AI painting engine?

@Gargron I also found it very unhelpful for dealing with telemarketers. 😁

@Gargron I mean if you're using it as a source of truth it's obviously really bad, but it's a text generator. It's an extremely impressive text generator.

@Gargron I disagree. For an ML model to be able to generate prose on essentially any topic under the sun that can reliably pass as human-authored? That's incredibly impressive! 'Useful' on the other hand . . .

@Gargron That's kinda what I was thinking! Thank you for confirming!

@Gargron You can as also about sources from #ChatGPT :)

@Gargron i havn't give it a try yet, seems like it can do lots of people work

@Gargron oh i like this take

@Gargron Inlove how this has 10 answers but no favorites ha!

@Gargron

An unreliable search engine is as good as it gets, I guess, if it's "a potential absolute limit on the whole practice of language modeling."

https://www.zdnet.com/article/openais-gigantic-gpt-3-hints-at-the-limits-of-language-models-for-ai/

@Gargron prediction: people will start to worship bots like they worship Muskrat. #ElonMusk

@Gargron as opposed to human "experts" who never get things wrong?

@Gargron it's like using Google's "I feel lucky" button, and then forcing yourself to use whatever answer it gives.

@Gargron I tend to disagree. No search engine I'm aware of comes up with a Monty Python Sketch about Mastodon and Twitter:

https://chaos.social/@rstockm/109456782032193274

@Gargron what is gpt ? 🤔

@Gargron I think you might be missing the point of it, it's not supposed to be search engine.

@Gargron more than that for me. It helped me (and saved many many many hours) by creating initial unit tests and documentation for my code.

@Gargron Exactly. It's bound to have an agenda more overarching than just interpreting and helping to reduce man hours.

@Gargron also very efficient for social occasions where you need some perfunctory poetry

@Gargron For some queries, I ask #ChatGPT for references to be completely sure of the results. I feel that this should be the default feature. But then large language models do not work that way.

@Gargron if you write code, you probably look for answers on stack overflow. When you find something that looks fine, you try it. The same applies for #chatgpt, but it takes a few seconds to have an answer instead of digging. Moreover, if you post a question on SO, you will probably be answered badly, verbally assaulted; it just does not happen with chatgpt.

@Gargron Reminds me of Cuil Theory.

http://cuiltheory.wikidot.com/what-is-cuil-theory

@Gargron This is so you can approach the result with an open mind. 😌

I thought the source was entire Reddit, StackExchange and Wikipedia.
@Gargron

@Gargron And when #gpt3 fails, it fails in a _really_, _very_ _awful_ way(s). Most obvious are the responses where output is plain non sense. They model is only a good as the data fed to it, and sometimes, even that isn’t sufficient. It’s a fun advancement for sure. More work remains to be done.

@Gargron
Someone needs to teach it how to cite its sources.

@Gargron wonkypedia?