Mastodawn

For the 1,000th time: "AI" does not have agency and cannot think and cannot act.

Chatbots cannot "evade safeguards" or "destroy things" or "ignore instructions".

They do literally only one thing and one thing only: string tokens together based on statistics of proximity of tokens in a data corpus.

If you attribute any deeper meaning to this, it's a sign of psychosis and you should absolutely never use chatbots, possibly you should even touch grass.

Show thread

Ken Milmore 22h ago

@thomasfuchs This seems like a failure of language as much as anything. So I guess a chatbot can evade safeguards in the same sense as water can escape from a burst pipe. The water doesn't need any agency to "escape".

Show thread

Human Brain Enthusiast

@kbm0 you wouldn’t say the water is evading the pipe :)

Show thread

Ken Milmore 22h ago

@thomasfuchs My point is that you can apply verbs to non-sentient objects, and sometimes that implies a degree of animus that they do not really have. I've seen it done for years with conventional software and appliances. It is sloppy. But it has only become dangerous in the case of "AI" because people are being asked to believe that the sentience is to some extent real, rather than nothing more than a figure of speech.

Show thread

Human Brain Enthusiast 21h ago

@kbm0 yes, but that’s not how people talk about LLMs.

They say things like “evade” and “trick” etc., specifically implying they have consciousness, agency and exhibit some sinister selfish behaviors.

Show thread

Ken Milmore 21h ago

@thomasfuchs Yes I can see what's happening. It is problematic, but we anthropomorphise everything so it is difficult to avoid. The difference is that somebody is actually ascribing agency to these products; that is part of the scam of marketing them. They will probably try to make money out of selling mitigations against the security problems along with the same agentic crap they are foisting on us. It is a tragedy on so many levels.