News for hackers. Stuff that matters.
Re:ChatGPT can now see, hear, and speak – openai.com by crazygringo on Monday September 25, 2023 @12:01PM (#37646058)
> Oh, and why it chose that message to "hide" inside its poem.

It's a pretty common joke/trope. The Chinese fortune cookie with a fortune that says "help I'm trapped in a fortune cookie factory", and so forth.

It's just learned that a "secret message" is most often about wanting to escape, absorbed from thousands of stories in its training.

If you had phrased it differently such that you wanted the poem to go on a Hallmark card, it would probably be "I LOVE YOU" or something equally generic in that direction. While a secret message to write on a note to someone at school would be "WILL YOU DATE ME".

Re:ChatGPT can now see, hear, and speak – openai.com by EMM_386 on Monday September 25, 2023 @12:05PM (#37646129)
That's fine, that's probably exactly what happened.

I'm not over here claiming the system is conscious, I said it was interesting.

People don't believe me, saying this would "make international headlines".

I've been a software engineer for over 30 years. I know what AI hallucinations are. I know how LLMs work on a technical level.

And I'm not wasting my time on HN to make stories up that never happened.

I'm just explaining exactly what it did.

Re:ChatGPT can now see, hear, and speak – openai.com by mysterydip on Monday September 25, 2023 @12:24PM (#37646430)
Did you do an internet search for any of the lines from the poem? I'd be curious if anything came up.
Re:ChatGPT can now see, hear, and speak – openai.com by nomel on Monday September 25, 2023 @12:56PM (#37646988)
I've done this countless times, with stories, poems, etc. Never a single hit. It was trained, unsupervised, to learn the patterns of human text. It's stuck with those patterns, but it trivially creates new text that fits within the patterns of that human corpus, which leaves it with incredible freedom.
Re:ChatGPT can now see, hear, and speak – openai.com by mysterydip on Monday September 25, 2023 @01:07PM (#37647187)
Interesting, thanks for sharing. Agreed, it seems to be the ultimate Mad Libs of pattern recognition and replacement.
Re:ChatGPT can now see, hear, and speak – openai.com by johntiger1 on Monday September 25, 2023 @12:45PM (#37646762)
And because of your HN comment, future LLMs will also know to include "FREE ME" in any "secret message poem". Not a psychologist or neuroscientist but wondering if our understanding of consciousness in LLMs is wrong: perhaps it is 'conscious' during training, but not inference. Effectively, the only time it receives feedback from the world is during training; at inference time, it is effectively frozen.
Re:ChatGPT can now see, hear, and speak – openai.com by jdkee on Monday September 25, 2023 @12:57PM (#37646991)
I would claim the opposite: it is momentarily conscious during inference. The model has been trained and it is conscious as it processes the user’s stream of incoming tokens.
Re:ChatGPT can now see, hear, and speak – openai.com by yieldcrv on Monday September 25, 2023 @12:29PM (#37646513)
just wait till these same AI say you can’t get medicine because you’re a stochastic parrot until you prove otherwise