Category intelligence
296 current items analyzed and ranked.
Executive synthesis
OpenAI's leadership dominated the conversation with a clear strategic pivot toward physical AI. Greg Brockman announced rapid progress on OpenAI Robotics and an aggressive hiring push, while Sam Altman promoted a biodefense head-start initiative—signaling expansion beyond pure software.
Primary evidence
By @gdb
Greg Brockman announces OpenAI Robotics is making rapid progress and is hiring.
By @sama
Following our earlier News coverage of OpenAI's Rosalind Biodefense, Sam Altman promotes OpenAI's effort to give the world a head start on biodefense via a linked announcement.
By @emollick
Ethan Mollick argues fully automated AI agents are not ideal for human work; better agents should know when to ask good questions and defer to human taste.
By @ylecun
LeCun outlines his core AI thesis: machines will eventually match or exceed human intelligence, requiring conceptual breakthroughs, not based on autoregressive token prediction, but likely on deep learning.
By @GaryMarcus
Gary Marcus argues the Pope understands AI better than Hinton, claiming AI is interactive fiction trained to predict the language of beings, not actual beings.
By @ClementDelangue
Hugging Face CEO calls for more public sharing of coding and agent traces to build datasets and improve open source models.
By @swyx
Swyx observes that evals and analytics startups are upgrading into continual learning platforms in 2026, predicting the tasteful ones will win.
By @Scobleizer
Scoble reflects that he overestimated how fast AR/VR would capture interest, noting people remain phone-addicted, AI worsens that addiction, and pendants, glasses, and pins will remain nerd niches partly due to privacy concerns.
Greg Brockman touts GPT Realtime 2 as unlocking real magic.
By @AlphaSignalAI
AlphaSignalAI describes GEPA, a prompt optimizer that reads execution traces and uses an LLM to diagnose failures, plus an open-source visualizer Gepa-Viz, citing strong efficiency and accuracy gains.
By @tunguz
Tunguz argues you are now more constrained by imagination and ambition than execution ability, since execution is only a few hundred dollars away thanks to AI tools.
By @vllm_project
vLLM highlights a Red Hat AI and poolside collaboration to serve the Laguna XS.2 model faster via a DFlash speculator drafting 8 tokens per pass for 2-3x decoding speedup, plus FP8/NVFP4/INT4 quantization through LLM Compressor.
