AI Reading for Tuesday April 15
OpenAI launches GPT-4.1, says it's better and 26% cheaper than 4o - OpenAI
A lot of the benchmarks still show o3-mini reasoning model as the head-to-head winner. Will be interesting to see if it beats GPT-4o and the current leader Gemini 2.5 in LM Arena, and if they release o3 soon.
One thing that was notable was how far Llama-4 Maverick dropped when they used the stock config instead of the LM Arena-optimized one. Kind of shows the limitations of trying to benchmark these super complex beasts in any single dimension, Goodhart’s law, etc.
GPT-4.1 didn't challenge Gemini 2.5 on coding, reasoning o3-mini is still the best OpenAI model. - aider
OpenAI plans to phase out GPT-4.5 from its API in July, presumably because 4.1 is better and cheaper. - TechCrunch
Suggests it was always a bragging rights model, and even at the super high price it cost too much to run.
Former OpenAI Employee Rages Against Sam Altman: "Person of Low Integrity" - Futurism
Kling launches 2.0 text-to-video model - YouTube
Some user experiments - Reddit
Joseph Stiglitz Warns Of The Looming Inequality Amid AI Monopoly Power - Forbes
What can workers do if they’re fired by AI? - FT
What ‘Ex Machina’ got right (and wrong) about AI, 10 years later - Fast Company
Meetings Won’t Be the Same When the Boss Sends an AI Bot - Bloomberg
Apple plans to collect user data about AI usage with differential privacy - The Verge
x
‘She helps cheer me up’: People form relationships with AI chatbots - the Guardian
AI-enabled cameras help blind people navigate
A weird phrase 'vegetative electron microscopy' tunneled into its way into scientific literature after a glitch in a pdf scan - The Conversation
DolphinGemma: How Google AI is helping decode dolphin communication - Google
This exhibit is closed temporarily for maintenance - Reddit
Follow the latest AI headlines via SkynetAndChill.com on Bluesky