AI Reading for Friday August 30
Apple, Nvidia in talks to invest in OpenAI’s latest funding round. - WSJ
OpenAI says ChatGPT has 200m WAUs, similar to Netflix - Cointelegraph
Dell increased server sales 80% on AI demand. - TechRadar
Seems like a big increase, AI isn’t 40% of total server demand across the board, maybe regular product cycle stuff and not just AI.
Big Tech Is Very Afraid of a Very Modest AI Safety Bill - The Nation
Google will let you chat with your emails. - TechCrunch
Chat with your Excel worksheet - CNET
KANs so hot right now - MIT Technology Review
How does the thing LLMs model differ from human language? Well, for one thing it’s trained to avoid certain modes of human expression, like being less racist, profane, and undignified than your typical human. Human language takes place on many levels including facial expressions and hand and body language, and evolves as part of a complex system of social interactions and internal processes which AI training doesn’t really access in its training corpus.
There is an element here of economists seeing something that works in practice and asking if it works in theory. I think LLMs have given the lie to some academic linguistics like Chomsky’s universal grammar theory. If humans had built-in lexers and parsers, human language would be more consistent probably, and more similar to computer languages than to LLM behavior.
Please do not use Tom Hanks in your evil machinations - Mashable
YiVal is an automated prompt evaluation assistant and improver. - GitHub
Follow the latest AI headlines via SkynetAndChill.com on Bluesky