2 ms·
This deserves wider distribution. It needs saying constantly. Overattribution is joining my discourse on this. All the language chosen by ai proponents favours
by ggm 3y ago
This deserves wider distribution. It needs saying constantly. Overattribution is joining my discourse on this. All the language chosen by ai proponents favours belief on a mind. Choice, says, infers, states, hallucinated, thinks.. these are crap words to use describing the systems behaviour of an llm or gpt.
- dkjaudyeqooe 3y agoI feel like infer is ok because it's a logical (mathematical) process not specific to humans.
- ggm 3y agoYes, I use it myself, but I do think it has potential problems. Brains make inferences. Self policing language critiquing AI is really important I think because analogistic thinking emerges when you try to describe something like machine generated texts, writing has always been predominantly intentional from a person or group, it has thought behind it. To emerge into a place where simple markov chain text streams have been superseded by LLM which can expose much more complex information makes it really hard. NLP is extracting information from the primer/prompt, the LLM is finding data matching the semantic intent of the primer text, the result(s) are marshalled into a cohesive whole. It's impressive Jim but it's not life as we know it. I sometimes wonder to what extent clippy defined the expectations here. The idea of a software "agent" has been around for a while. If clippy had state, and had been coded to respond with "I see YET AGAIN you are ..." people might have been freaked out a bit earlier.