Stop AI, Pause AI and Safety AI people need to read and share this. General sceptics and realists too. An honest inventory of the real reactions of disappointment to GPT5. Not the Wowslop of the paid mainstream shills and the suspicious silence of the AI doomers.
Thanks to @xriskology for this quote by Adam Becker. "ChatGPT is a text generation engine that speaks in the smeared-out voice of the internet as a whole. All it knows how to do is emulate that voice, and all it cares about is getting the voice right. In that sense, it’s not
@MrEwanMorrison I'm sure you saw that I quoted you, too. :-)
@xriskology Bingo! Yes, I'm very glad that we're on this and seeing the importance of it. Reality checks. Nothing stops the tech accelerationists faster than hitting the brick wall of reality.
@MrEwanMorrison @xriskology Phenomenal quote, albeit depressing!
@MrEwanMorrison @xriskology There is an implication in that quote that if ChatGPT did have a "notion of truth or falsehood", if it didn't hallucinate, then it would be "ok". Setting aside what is "truth", curious as to what you and @xriskology might think of that. Happy to be referred if you've previously
@narrascaping @xriskology It's always hard to eradicate all trace of anthropomorphism from the language around AI. For example, it's of great urgency that we find a good strong single word to replace "hallucination", which actually explains the process of algorithmic token groupings.
@MrEwanMorrison @xriskology "It's not making a mistake when it hallucinates." Because so-called "hallucinations" are only defined subjectively by humans. From a probabilistic point of view, all LLM outputs are correct. You wouldn't say that a slot machine is faulty just because it didn't hit "777".