On 12.08.26 18:32, Matthias Geiger wrote:
Furthermore, it is simply not possible to build an LLM that does not hallucinate [0].

I have two problems with this statement.

A: This ballot is not about LLMs. It is about AI, which is a strict superset. You can (and some people already do) build non- or not-entirely-LLM-powered AI systems.

B: Hallucinations are a strict subset of the nonsense LLM/AI input can introduce to software. This is why IIRC we more-or-less agreed that *if* we allow LLM input to Debian, any such nonsense is the responsibility of the DD in question, who is thus obliged to check for it *before* submitting or updating the output in question.

--
-- regards
--
-- Matthias Urlichs

Attachment: OpenPGP_signature.asc
Description: OpenPGP digital signature

Reply via email to