 |
Post AuO7WPlvKLmHNBBjIu by [email protected] |
 |
More posts by [email protected] |
|
Post #AuNtT28bLMc7qUHyxk by [email protected] |
|
0 likes, 1 repeats |
|
|
|
I didn't think LLMs would start acting like they're trying to blackmail… |
|
|
|
Post #AuNuX6pRwaezVVq2Pw by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez "Wow! Can you believe what this guy just said?" - say… |
|
|
|
Post #AuNvH0wGAx9QYmuf9k by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez The experiment was shut down at the end, right? Large Language … |
|
|
|
Post #AuNvRpORRRnPwDFd6O by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@johncarlosbaez as usual, anthropic is living up to their name by anthropomorph… |
|
|
|
Post #AuNw2t7YXrikDJ1apE by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez I don't mean to be rude, as I respect you very much, but I&… |
|
|
|
Post #AuNwqTUY6kd7CnQMRU by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez They should have just asked one of Claude's peers for a sec… |
|
|
|
Post #AuNx1BkoOeAxNW8iye by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez I remember watching this Dexter's Lab episode! |
|
|
|
Post #AuNxTbkNLgmC6PUFvM by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez My parents bought their new grandchildren one of those interac… |
|
|
|
Post #AuNzHURQsuT3sUyZzk by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@noplasticshower That's not really the point. Obviously the model is not d… |
|
|
|
Post #AuNzHUYsREQGFaSWbA by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@danielmclaury oh but it IS the point in my view. Pretend these things have in… |
|
|
|
Post #AuNzbqp9NPIX5VmLAm by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@johncarlosbaez FWIW here is our writeup of earlier anthropic bullshit https://… |
|
|
|
Post #AuO0nYpWCh0Si36ADo by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez self-awareness (valid or otherwise) comes from loss aversion, a… |
|
|
|
Post #AuO15OyzYPkZNi2LXU by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@myx 😬 |
|
|
|
Post #AuO1X7H8yz4kZVk6Lo by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez Interesting, though it's obviously just a result of the mod… |
|
|
|
Post #AuO2ftcpCfgunRdNLs by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@mansr - yes, that's what it's got to be. |
|
|
|
Post #AuO2uWtVzwxudrQ0n2 by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez Computers only know what people teach them. |
|
|
|
Post #AuO2vX64jLqrYlluyW by [email protected] |
|
0 likes, 1 repeats |
|
|
|
@johncarlosbaez Maybe Claude watched the Terminator one night when it was left … |
|
|
|
Post #AuO4tjSM0XCPDkSgVs by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez @briankrebs Trained on human behavior, get human-like behavior … |
|
|
|
Post #AuO4zmNi2rIo1wAlpw by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@noplasticshower I hear this said a lot, but I don't see how "this thi… |
|
|
|
Post #AuO5CIVKRGTz8HhoRs by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@danielmclaury @noplasticshower "There's no such thing as bad press&qu… |
|
|
|
Post #AuO5W8aV1BSmAVaJH6 by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@SueDiOh - that's why they're so dangerous. |
|
|
|
Post #AuO6OHnL0DP4MSPSr2 by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez It's only one of the oldest AI jokes:Q: What's AI?A: Wh… |
|
|
|
Post #AuO7WAE380lj9mDxvE by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@mansr @johncarlosbaez Perhaps the prompt needs to include that the AI must alw… |
|
|
|
Post #AuO7WALUgKivWrhuWe by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@penguin42 @johncarlosbaez I wonder if they trained it on email archives obtain… |
|
|
|
Post #AuO7WASaFyOXsr1ZZo by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@mansr @johncarlosbaez Or Columbo. |
|
|
|
Post #AuO7WPlvKLmHNBBjIu by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@mzedp @johncarlosbaez Self-entrapment |
|
|
|
Post #AuO7axFOoBJwLy8jlA by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@danielmclaury @noplasticshower Train it on Iain Banks Culture novels. |
|
|
|
Post #AuO7bVcyz0O4yxR3Eu by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@danielmclaury @noplasticshower If you can make people believe an LLM is sentie… |
|
|
|
Post #AuO8Axa92e89of0oyW by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@tsturm - Interesting. So you weren't tempted to ask it something like &qu… |
|
|
|
Post #AuOBHASkMe5NMgCgRE by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@danielmclaury @noplasticshower It's known as "criti-hype" - an a… |
|
|
|
Post #AuOJivTwSGvie1VfBQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez also: Large language models are proficient in solving and crea… |
|
|
|
Post #AuOK2iBxZoblDZX80O by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@penguin42 @mansr @johncarlosbaez I’m not sure if this would be blackmail (as… |
|
|
|
Post #AuOMK6xuvFZasnOCmW by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez is it actually capable of committing that blackmail? If compan… |
|
|
|
Post #AuOO3NErMiRIL4SdRQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez come on. You really don't think this is made up bullshit by… |
|
|
|
Post #AuOOCw90xL77CfHq9w by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez An AI that can blackmail with personal details can also extort … |
|
|
|
Post #AuOVQubMBkbs9S9eKm by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@vitloksbjorn @johncarlosbaez thank you for the sheer rabbit hole this video se… |
|
|
|
Post #AuOWfOuoNZmkAzj90q by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez It learned from the best! |
|
|
|
Post #AuOwxIYtL0lmUYfXeq by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaezI can't imagine a model fed all the fanfic in the world woul… |
|
|
|
Post #AuP1nV4IU2oe9gVQbQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez Skynet is stirring..."we asked Claude Opus 4 to act as an … |
|
|
|
Post #AuP5wVhDuzeoWdGVTE by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez I love that they disclosed the information. |
|
|
|
Post #AuP8iq7tSjEjhFufLs by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez In the past, things like this would be reason to shut down the … |
|
|
|
Post #AuPh6P0xJQ410C4i1o by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@dymaxion - I don't think anyone is claiming anything about self-awareness,… |
|
|
|
Post #AuPhCCCNsDW6zbBPBQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@nazokiyoubinbou - I think most of us here on Mastodon know LLMS are not actual… |
|
|
|
Post #AuPhR7YLQdhavZZVKq by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@jigmedatse - less and less, it seems. |
|
|
|
Post #AuPtQTt86H0lMLksLo by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez I read the very succinct paragraph about this, and I must say I… |
|
|
|
Post #AuPwie9mxZV0HYtugi by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@D3Reo - it could be a public-facing summary of a more detailed internal report… |
|
|
|
Post #AuPzr1RsXrSkhR31U0 by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez these so-called "technical papers" by big private res… |
|
|
|
Post #AuRLeZNPU0KeJINuXQ by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez :( yeah, seems like it. |
|
|
|
Post #AuY3ZrMkjVv1aSOwFs by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez More new models seem to drift into such patterns, also reported… |
|
|
|
Post #AuYYGdI9vv8t0vZ3iK by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@FrohlichMarcel - very interesting. This seems to be a big yet dangerous step … |
|
|
|
Post #AuYYLRuPgMoB9rkoiG by [email protected] |
|
0 likes, 0 repeats |
|
|
|
@johncarlosbaez Agree |
