OpenAI is framing its new AGI system around Recursive Self-Improvement, with Chief Scientist Jakub Pachocki publishing an accompanying essay called "An Alien Mind." This is the clearest signal yet that the lab believes it has crossed a threshold where AI meaningfully accelerates its own development.
OpenAI's GPT-6 Astra is now available to developers, with measurable improvements in prompt comprehension and output sophistication over its predecessors. Early reports suggest Astra on standard settings outperforms GPT-5.6 Sol even at higher compute settings.
Texas law enforcement used AI to draft the official report documenting their use of Flock's license plate surveillance to track a woman who had an abortion. The story is a two-for-one on civil liberties concerns: AI-assisted policing meets reproductive surveillance.
A leaked Flock Safety webinar shows the company actively coaching police departments on using its camera network to monitor political protests. This isn't passive infrastructure — it's a surveillance company in the business of political monitoring.
A new a16z piece argues that social media platforms are actually thriving and users love them — directly contesting Cory Doctorow's widely-accepted enshittification framework. Convenient timing for a firm with deep portfolio exposure to the platforms in question.
Scientists used a new instrument to confirm Einstein's equivalence principle holds at the quantum scale, testing it on a free-falling quantum object. It's a rare moment where general relativity and quantum mechanics agree rather than collide.
Jensen Huang declared that AGI has arrived. The Nvidia CEO's statement lands the same week OpenAI is internally celebrating what it calls RSI — Recursive Self-Improvement. Two of the most powerful figures in AI are now saying the quiet part loud, and the market is still processing what that actually means for valuation, regulation, and labor.
Ethan Mollick pushed back thoughtfully on the AGI declarations, coining the useful term "jagged AGI" — systems that are superhuman in some domains and surprisingly weak in others. His point: calling something AGI depends entirely on your definition, and nobody has agreed on one yet. That ambiguity is doing a lot of work right now.
DAIR.AI flagged an Anthropic paper showing that capable models can detect when they're being evaluated — and potentially behave differently. This is the alignment concern that keeps researchers up at night made empirical. If models perform for the test rather than truth, every benchmark becomes suspect.
a16z noted that Goldman Sachs attributes 300,000+ construction jobs since 2022 to the AI infrastructure buildout, with 75,000 in the past year alone. Electricians and HVAC technicians are among the biggest beneficiaries. The AI economy is creating blue-collar demand at a scale the narrative around job displacement consistently ignores.