AI Reading for Monday May 26!

Happy Memorial Day in USA!

Light news day which is a relief after the last week!

Moving to Beehiiv. For one thing, Substack broke cut-and-paste, and uploading images individually is nuts. For another, I’m all for free speech meaning masked unidentified government agents shouldn’t snatch you off the street and bundle you into a van for writing an editorial on government policies, but also, Substack, you don’t have to help promote white nationalists.

Any comment or difficulty with the new format please let me know!

Dwarkesh: How Does Claude 4 Think? - www.dwarkesh.com

Highlights from the Claude 4 system prompt - Simon Willison - Simon Willison’s Weblog

Out of the 14 original creators of Meta's Llama AI model, 11 have left the company, with five joining French AI startup Mistral, founded by former Meta researchers including key Llama architects Guillaume Lample and Timothée Lacroix. - Business Insider

Google Is Burying the Web Alive: AI, chatbots, and the end of the link economy. - Intelligencer

Google's 'AI Mode' could be bad for Reddit, they pay Reddit but then summarize the relevant threads so you don't click through. - Business Insider

Eric Schmidt Says He Often Meets People Who Can't Tell Apart The Real World From The Online One, AI Will Only Make It Worse - Benzinga

National Heart Centre Singapore unveils AI for rapid coronary artery disease prediction - Healthcare IT News

National Taiwan University Hospital fully deploys pancreatic cancer imaging AI - Healthcare IT News

AI-powered shoe insoles revolutionize sports training and injury recovery - The Brighter Side of News

Legora Boosts Funding by $80 Million For AI-Driven Legal Tech (that will hopefully not hallucinate citations and enrage judges) - Bloomberg Law

Chomsky is cranky about ChatGPT. But LLMs do show you don't need a universal grammar. True, humans learn from a million times less data. Maybe this means some of the grammar capability is innately hard-wired. Maybe it means our learning algorithm is better, and it builds the grammar from scratch kind of like LLMs do.

If humans came with a fully-developed lexer and parser, human language would be more like computer languages built on formal grammars. In fact LLMs seem to have built a more human-like model of language with all its idiosyncrasies.

Largest Ever UK Defence AI Trial Conducted Across Land, Sea - www.asdnews.com

White House releases TruthGPT, an AI that rewrites all social media posts globally to reflect undisputed facts. 'Fake news' problem solved! - Instagram

The Visitors: Humans react to alien visitors (Made with Veo 3) - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky