AI Reading for Monday July 7

The Hollywood thing feels a little too good to fact-check, would wait til people replicate it with disclosed, non-leading prompts. Grok blaming Elon for the flood in Texas? It’s all mixed up. Time for Grok to get some professional help

Labeling beats working the fields though?

Using AI to help start tough conversations and even ask fellow humans for help.

Smart guy, saw him speak. But oddly Google is failing at popularizing their very good AI ecosystem. This stuff is hard. Despite ChatGPT having a lot of product shortcomings, everybody thinking GPT-4o must be better than o3 because it’s a higher number, Google and others catching up on base models, everybody prefers using ChatGPT over Copilot and Google.

Agents and robots can learn by rote imitation, or by searching a well-modeled solution space (like chess or Go), or by trial-and-error (reinforcement learning). RL can be very effective with a proper reward objective, but takes good science, data and compute.

But if you have to replay points due to obvious errors, it’s not good.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky