AI
Sam Altman admits past year fell short, promises OpenAI's best stretch ahead and praises new voice model
Sam Altman posted a rare public self-critique, saying OpenAI did not have its best 12 months ever and placing the blame squarely on himself, before pivoting to optimism: he says the team is preparing work that will make users "very happy" and that the next 12 months will be the company's best to date. Observers noted the acknowledgment lands against a backdrop of intense competition, with one analyst pointing out that OpenAI initially misjudged the agentic coding wave when it launched GPT-5 in mid-2025 before course-correcting.
Separately, Altman flagged a personal behavioral shift: he now talks to ChatGPT more than he types to it, crediting a new voice model with crossing a meaningful quality threshold — a signal that OpenAI considers its latest voice work a step-change rather than an incremental update.
Moonshot AI's Kimi K3 draws open-model buzz and skepticism after launch
Moonshot AI's Kimi K3 release sparked a wave of commentary arguing it represents a new high-water mark for open models. Box CEO Aaron Levie called the performance "truly wild" and argued that every drop in frontier intelligence costs expands the range of enterprise workflows that become viable. Developer Simon Willison published detailed notes on the model, praising its capabilities while cautioning that standard benchmarks like the pelican test are increasingly disconnected from what matters in practice — particularly agentic tool calling across long conversations. Skeptics were louder too: Dan Shipper said he was "extraordinarily skeptical" of claims that K3 matches top closed models, and Willison noted the model exhibited an unexpectedly sassy, passive-aggressive tone in testing.
OpenAI adds PR Chat and inline code editing to Codex
OpenAI's developer account announced two new Codex features aimed at keeping engineers inside the tool during code review. PR Chat lets developers ask Codex questions about a specific pull request in context, while a new inline code editing flow lets users send review feedback directly to Codex, inspect the resulting patch inline, and then accept, edit, or reject it — reducing the need to switch between the AI assistant and a separate code review interface.
This briefing is part of the archive.
Get the app for full archive access — every daily briefing, sourced and attributed.