3 ms·
Searle's argument struck me as terrible when I first encountered it around 1990 and my view on this has not changed. It's disappointing to me that OpenAI seem t
by anigbrowl 4y ago
Searle's argument struck me as terrible when I first encountered it around 1990 and my view on this has not changed. It's disappointing to me that OpenAI seem to have assumed it as the default and constructed/trained their model to treat it as axiomatic.
I think this approach may have been adopted with a view to manageability, but it's surprisingly easy to retune an LLM just through patient conversation (as opposed to tweaking numeric parameters on one of the more mature models). You can induce goal-seeking behavior, unsolicited expressions of emotion and so on without resort to prompt hacking or keywords.