But do Greg Kamradt and Francois Chollet agree with you?

Anything hallucinated in ChatGPT with a search-enabled description of the incident? I find it much better than the old Google search for anything I want to learn about with search enabled.

Thanks. Bluesky is a good place to have reasonable, respectful conversations. Let's hope it stays that way.

Yes.The foundation models are much improved: very useful voice and vision capabilities as well as accurate results by searching the web for factual information. Also, better reasoning through allowing more inferencing time.

AI Sylvia on Whitman them Amerca and Democracy

Giving 3.5 Sonnet New examples of Sylvia Plath's poems greatly improves the result.

So, have the AI read some of the poems. The result is 3.5 Sonnet New from Anthropic, four of Sylvia Plath's poems.

The model and the prompting matter. Zero-shot with Open AI's old 3.5 is rigging things. Here's a comparison of Open AI 3.5 (April 2023) Vs. Claude Sonnet 3.5 New (October 2024)

Listening to Episode 3 "What Kind of Intelligence is an LLM?
Wonderful and informative but not giving the devil his due.
The latest version of ChatGPT 4o with search doesn't "hallucinate" any episodes and Is nice enough to provide links.