Megan Morrone
@megmorrone.bsky.social
Journalist @axios. Writing about how AI will save/destroy the future. I live in San Francisco, but mostly I live here, on the internet. Previously: Fast Company, BBC, Protocol, Medium, Daily Tech News Show, This Week in Tech, TechTV.
AI companies could reduce hallucinations, but it would make their models slower, costlier, and less “helpful.”
So they optimize for speed and scale instead.
That’s why chatbots still confidently make stuff up. And probably always will.
So they optimize for speed and scale instead.
That’s why chatbots still confidently make stuff up. And probably always will.
June 4, 2025 at 1:02 PM
AI companies could reduce hallucinations, but it would make their models slower, costlier, and less “helpful.”
So they optimize for speed and scale instead.
That’s why chatbots still confidently make stuff up. And probably always will.
So they optimize for speed and scale instead.
That’s why chatbots still confidently make stuff up. And probably always will.
Some highlights:
– RAG and SAFE aim to reduce hallucinations, but can’t eliminate them
– Newer models hallucinate more because they reason more
– “Hallucination-free AI may never arrive”
– RAG and SAFE aim to reduce hallucinations, but can’t eliminate them
– Newer models hallucinate more because they reason more
– “Hallucination-free AI may never arrive”
June 4, 2025 at 1:02 PM
Some highlights:
– RAG and SAFE aim to reduce hallucinations, but can’t eliminate them
– Newer models hallucinate more because they reason more
– “Hallucination-free AI may never arrive”
– RAG and SAFE aim to reduce hallucinations, but can’t eliminate them
– Newer models hallucinate more because they reason more
– “Hallucination-free AI may never arrive”