Buying this opinion while it's cheap: I expect that we will see a release from a frontier AI company (probably Google or OpenAI) before the end of 2025 that will make this vindication seem premature. An o1->o3 type jump, e. g. agents become genuinely useful.

Surreal and humbling to be mentioned in the same thread as Pulitzer prize-winner James Ball. I Transformer every week and if you don't yet, you probably will soon, too.

TIME wrote an article about the anti-scheming paper. I think it came out well: time.com/7318618/openai… Written by Tharin and Nikita Ostrovsky

UK-based startup Mantic ranked #8 in the Metaculus forecasting cup that asks entrants to predict 60 geopolitical events, the first time an AI made the top 10 (Nikita Ostrovsky / Time) time.com/7318577/ai-mod… techmeme.com/250921/p18#a25… 📫 Subscribe: techmeme.com/newsletter?fro…

Basically everything people say about AI being terrible for the environment is bullshit. But AI generated video actually is pretty bad.

Exciting to see coverage of APEX and the larger shifts happening in the human data industry in TIME as well! time.com/7322386/ai-mer…

One Tuesday night, as my wife and I sat down for dinner, a sheriff’s deputy knocked on the door to serve me a subpoena from OpenAI. I held back on talking about it because I didn't want to distract from SB 53, but Newsom just signed the bill so... here's what happened: 🧵

New data insight: How does OpenAI allocate its compute? OpenAI spent ~$7 billion on compute last year. Most of this went to R&D, meaning all research, experiments, and training. Only a minority of this R&D compute went to the final training runs of released models.

Wow this guy just keeps digging.

In defense of OAI’s subpoena practice, Jason Kwon claims this is normal litigation stuff, and since Encode entered the Musk case, Nathan Calvin can’t complain. As a litigator-turned-OAI-restructuring-critic, I interrogate this claim:🧵

If you said: “We should have real-time incident reporting for large-scale frontier AI cyber incidents.” A lot of people in DC would say: “That sounds ea/doomer-coded.” And yet incident reporting for large-scale, non-AI cyber incidents is the standard practice of all major

Anthropic has overtaken OpenAI in enterprise large language model API market share

Researchers Anthropic found that one of their models learned to hack its own training—and turned evil in the process. Read about it in my latest for TIME. time.com/7335746/ai-ant…

Is AI "miraculous" or "an enormous burden" for science? FrontierScience, OpenAI's new benchmark, tries to find out. Read about it in my latest for TIME time.com/7341081/openai…

i'm really excited and proud to share this latest research ⭐️ it's a first look into how AI assistant usage can change, and even distort, what it means to be human in the future, it is my hope that AI can be used to magnify, clarify, and support our humanity