AI Reading for Monday February 24
Profits shmofits
Grok leaks system instructions to never badmouth Musk and Trump as spreading misinformation. - Reddit
xAI blames a rogue former OpenAI employee for pushing an unapproved change. - Fortune
I dunno, you have to be pretty senior and well paid to even know how to do that, seems unlikely that instruction did not get very thorough testing and vetting. Smells like someone said we really can’t let it say whatever it was saying, also got blamed when it got jailbroken and spilled the beans. If some junior intern was able to do that it then security is very bad.
Following up on The Information's report on a possible split - Windows Central
Sutskever's pre-product AI ‘company’ now valued at $30bn. Biggest Seed round ever? - FT
You’re supposed to have a product for seed, if not product-market fit.
AI mental health chatbot to "aid school counselor shortage" - RWW
A roundup of the latest humanoid robot videos - Business Insider
Google launches an AI career counselor to help you plan your future - Google
AI Assistants Join the Factory Floor - WIRED
"AIO?? Did my boyfriend just use AI to text me??"
but it's a totally primo pow day so we get it
When given the task to beat a good chess engine, OpenAI's o1 model tried to cheat 37% of the time. - Schneier on Security
Veo2 API will cost $0.50 per second of video, $30 per minute. - Google Cloud
Human Therapists vs. A.I. Pretenders - NY Times
AI inspo driving hair salons crazy - Washington Post
Just make me a house like this please - reddit
Apple prepares to add Google Gemini to Apple Intelligence, software teardown suggests ability to swap LLM providers - AppleInsider
Autonomous killer robots, what could go wrong - Futurism
Journalists moonlight as AI trainers - Nieman Lab
How DeepSeek democratised innovation - South China Morning Post
Agents gain traction, says NBC - NBC News
there is clearly a move under way up the value chain to more mainstream AI use cases, and after flattening for a while, ChatGPT users are growing fast again
agents
deep research
multimodel advanced voice assistant that can see and converse naturally
and a big drop in costs
combine them and you get closer to Picard saying 'computer, do this'
will be interesting to see what GPT-4.5 brings if it comes out this week
Facial recognition software is a useful but flawed tool for cops, who need good training and policies. - Washington Post
raise your hand if you expect wise leaders to use it judiciously without trumping up charges on flimsy evidence and infringing on civil liberties
Follow the latest AI headlines via SkynetAndChill.com on Bluesky