On 4/3/25 02:41, Adrian Godwin via cctalk wrote:
AIs that are wise rather than idiot-savants ought to be useful, in the way sci-fi writers imagine. However, statistical systems can't surpass their training, and their training contains flaws - many, in the case of uncurated internet grabs. And the compulsion to not admit to a lack of knowledge and substiture - perhaps unknowingly - hallucinations is absurd. I cannot see any justification for that.
The justification for that compulsion is fairly obvious if you take "do no harm" out of the equation. An AI that says "I don't know" loses more business, aka profit, than an AI that spouts garbage.
AI and LLMs strengthen my belief that humanity's technology surpassed our ethics decades ago. Like, trampled any useful code of behavior into the dirt. I don't have the faintest idea what to do about that, either.
Doc
