This breathless two-minute paper 
<https://www.youtube.com/watch?v=bQE5CPxlBz0> references an arXiv preprint 
<https://arxiv.org/pdf/2303.12712.pdf> from Microsoft Research: Sparks of 
Artificial General Intelligence: Early experiments with GPT-4.

I highly recommend reading this paper. The researchers have taken Félix's 
suggestion to the next level and beyond.

The preprint is 155 pages long, consisting of prompts to GPT-4 and analyses 
of results. The paper highlights the strengths and limitations of GPT-4 in 
far more detail than popular accounts.

The sections on mathematics are particularly intriguing and mysterious. On 
the one hand, GPT-4 writes spectacularly well. I get a strong impression 
that its responses involve planning. But this impression seems to be 
misleading, as its math mistakes show. 


*Summary thoughts*

I do not understand why GPT-4 (or its predecessors) can create coherent 
text with the limitations revealed in the paper. The more I learn about GPT 
systems, the more mysterious they seem.

Nobody, including experts, can predict where all this will lead. Current AI 
research is true intelligent design :-)

Edward

-- 
You received this message because you are subscribed to the Google Groups 
"leo-editor" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/leo-editor/0795839c-8b49-4691-ae8a-097b00474bb1n%40googlegroups.com.

Reply via email to