4 ms·
Information wants to be free. The AI safety movement, while it appears on the surface to be a legitimate concern, has struck me so far to be populated with a c
by himinlomax 3y ago
Information wants to be free.
The AI safety movement, while it appears on the surface to be a legitimate concern, has struck me so far to be populated with a certain, shall we say, type of individuals of a certain anti-freedom persuasion.
Consider how Midjourney censors results to placate the horrendous Chinese dicatorship. They are not the good guys, not in any way.
- saurik 3y agoThe point of AI safety is not--and never was--"oh no the AI might say something offensive" or even "oh no the AI might answer a question that helps someone bad"; the people who have been staring at AI safety for decades now have always been staring at problems like "oh no the AI hates me [<- something Bing's AI already did once] and on purpose gave me a bad recipe suggestion that poisoned me" or "oh no the AI was given direct access to the factory and decided it would be more efficient to produce killer drones to steal stuff than a better product to sell".
- crooked-v 3y agoOr to put it another way, actual AI safety researchers/advocates don't really care about making 'harmless' or 'ethical' AIs in the sense that the LLM crew prattle about... what they care about is making 'sane' AIs that behave in ways that are intuitively sensible to average people. The censorship brigade with LLMs is something new and mostly unrelated that is doing basically surface-level nonsense to limit responses for milquetoast corporate culture reasons, because actually enforcing 'sane' behavior on an LLM is impossible with the current state of the art.