8 posts
OpenAI says AI has moved from minute-scale tasks to hour-scale work, with month-scale projects next. Its answer is a people-first industrial policy, not just more regulation.
OpenAI Codex manually sets Qwen3 transformer weights to solve 10-digit integer addition with only 343 parameters at 100% accuracy, passing 10 million random tests.
Frontier AI agents successfully exploited 72% of smart contract vulnerabilities end-to-end. Here's what OpenAI, Paradigm, and OtterSec's EVMbench found.
Watchdog group Midas Project alleges OpenAI's GPT-5.3-Codex was released without required high-risk model safeguards under California's SB 53 AI safety law. OpenAI pushes back, but safety researchers say the violation is clear-cut.
AI made you 10x more productive. Your salary didn't go up 10x.
The story of how GPT-5.2 discovered that a physics formula believed to be "zero" for decades was in fact nonzero, and even proposed a general formula.
A brutally honest warning from a 6-year AI startup founder who can no longer hold back.
Lessons learned from an internal OpenAI team that built and shipped a million-line software product in just 5 months using only Codex agents.