Category intelligence
462 current items analyzed and ranked.
Executive synthesis
OpenAI dominated the day's discourse with Sam Altman announcing GPT-5.3-Codex rolling out to Cursor, GitHub, and VS Code, alongside the milestone of 1M+ Codex App downloads in its first week. Altman framed 5.3 as a stepping stone—'not solved yet, but 5.3 will help build the thing that solves it'—while revealing cybersecurity concerns are gating the API rollout.
Primary evidence
Continuing our Social coverage of GPT-5.3 Codex, Sam Altman announces GPT-5.3-Codex is rolling out today in Cursor, GitHub, and VS Code.
By @sama
Sam Altman announces Codex App surpassed 1 million downloads in its first week with 60%+ weekly growth in overall Codex users. Commits to keeping Codex available to Free/Go users after the promotion, possibly with reduced limits.
By @OpenAI
Following earlier News coverage of the Anthropic-OpenAI ad battle, OpenAI announces starting to roll out ads in ChatGPT for a subset of US free and Go users. Ads are labeled as sponsored and visually separate from responses. States ads don't influence ChatGPT's answers.
Building on yesterday's Social coverage of GPT-5.3 Codex, Sam Altman states that while whatever is being discussed isn't 'solved yet,' GPT-5.3 'will help build the thing that solves it' — suggesting GPT-5.3 is a stepping stone toward more capable systems.
By @sama
Building on yesterday's Social coverage of GPT-5.3 Codex, Sam Altman explains the GPT-5.3-Codex API delay: it's their first model at a high bar for cybersecurity, and the extra safety work is taking longer than expected.
By @AravSrinivas
Arav Srinivas announces Perplexity has upgraded its Advanced Deep Research to use Claude Opus 4.6, claiming lead on Google's DSQA benchmark. Rolling out to Max users immediately, Pro users gradually.
By @simonwillison.net
Willison discusses HBR research showing that AI productivity boosts can lead to burnout and mental exhaustion, noting he's experienced this personally. Links to his blog post reflecting on the findings.
By @emollick
Mollick observes that faking continual learning and memory for AIs works better than expected, and predicts that if true continual learning is achieved, it will rapidly shift the AI capability frontier.
By @natolambert
Delivering on the Social teaser from yesterday, Nathan Lambert publishes detailed analysis of Opus 4.6 and Codex 5.3, noting model releases are harder to evaluate in 2026. Says he's in his 'post-benchmark era.' Claude still king but Codex is closer than ever.
By @fchollet
Chollet argues a true solution to intelligence must autonomously produce abstractions that compose well, stack well, and stand the test of time — and that no current technology, including gradient descent, achieves this.
As covered in News earlier this week, OpenAI posts 'You can just build things' with a link, likely related to their Super Bowl LX Codex ad campaign.
By @emollick
Ethan Mollick observes that writing satisfying medium-length stories remains harder for LLMs than math proofs, music generation, research reports, and code — noting the irony since they're 'language models'.
