korrents

On the map

Tap a claim on the ring to put it at the centre.

← Codex team's new features likely do not work well with models not trained on that mechanism

17 connected korrents · 16 moments on record from 17 Oct 2025 to 5 Sept 2026.

Everything filed under coding agents coding agents Everything filed under AI alignment AI alignment Everything filed under recursive self-improvement recursive self-improvement Everything filed under LLMs LLMs Everything filed under social media social media Everything filed under Anthropic Anthropic Everything filed under taste taste Everything filed under software quality software quality Same subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subject Read this korrent: Codex team's new features likely do not work well with models not trained on that mechanism Codex team's new features likely do notwork well with models not trained on thatmechanism Last stated 3 days ago 5 Sept 2026 MZ Mario Zechner — holds since 2026-09-05 — tap for who they are Same subject: The code today’s models and agents write is very hard to follow: you get better performance without knowing why, and no way to fix it when it breaks. — tap to centre the map on it The code today’s models and agentswrite is very hard to follow: youget better performance withoutknowing why, and no way to fix itwhen it breaks. Last stated 2 months ago 19 Jul 2026 ES Elizabeth Stone — holds since 2026-07-19 — tap for who they are Same subject: A saboteur among our AI investigators would be hard to spot, because these models are sloppy and spiky enough that a suspicious error just looks like ordinary incompetence. — tap to centre the map on it A saboteur among our AIinvestigators would be hard to spot,because these models are sloppy andspiky enough that a suspicious errorjust looks like ordinaryincompetence. Last stated a week ago 1 Sept 2026 AC Ajeya Cotra — holds since 2026-09-01 — tap for who they are Same subject: Coding models are worst at exactly the thing an AI research explosion would need: code that has never been written before. — tap to centre the map on it Coding models are worst at exactlythe thing an AI research explosionwould need: code that has never beenwritten before. Last stated 11 months ago 17 Oct 2025 AK Andrej Karpathy — holds since 2025-10-17 — tap for who they are Same subject: There is no real impediment to models running their own research loop and improving themselves at a far more rapid rate. — tap to centre the map on it There is no real impediment tomodels running their own researchloop and improving themselves at afar more rapid rate. Last stated a month ago 30 Jul 2026 JD Jeff Dean — holds since 2026-07-30 — tap for who they are Same subject: Newer model releases have got worse at producing text that is actually pleasant to read and understand. — tap to centre the map on it Newer model releases have got worseat producing text that is actuallypleasant to read and understand. Last stated 2 weeks ago 25 Aug 2026 AR Armin Ronacher — holds since 2026-08-25 — tap for who they are Same subject: Platforms will lose the ability to detect AI-generated content as the models improve, and should say how confident they are rather than pretend. — tap to centre the map on it Platforms will lose the ability todetect AI-generated content as themodels improve, and should say howconfident they are rather thanpretend. Last stated 2 months ago 9 Jul 2026 AM Adam Mosseri — holds since 2026-07-09 — tap for who they are Same subject: The interesting question is not what the next model will be able to do, but what the models already released can do that nobody has worked out yet. — tap to centre the map on it The interesting question is not whatthe next model will be able to do,but what the models already releasedcan do that nobody has worked outyet. Last stated 6 months ago 19 Mar 2026 SW Simon Willison — holds since 2026-03-19 — tap for who they are Same subject: Models have no research taste yet, which is why they complement researchers rather than replace them. — tap to centre the map on it Models have no research taste yet,which is why they complementresearchers rather than replacethem. Last stated 2 months ago 26 Jun 2026 NB Noam Brown — holds since 2026-06-26 — tap for who they are Same subject: A coding agent does not learn from its mistakes the way a person does -- it repeats the same error indefinitely unless a human notices and writes it down. — tap to centre the map on it A coding agent does not learn fromits mistakes the way a person does-- it repeats the same errorindefinitely unless a human noticesand writes it down. Last stated 5 months ago 25 Mar 2026 MZ Mario Zechner — holds since 2026-03-25 — tap for who they are Same subject: A coding-agent company should not train its own model: it has to stay neutral ground for models to compete on. — tap to centre the map on it A coding-agent company should nottrain its own model: it has to stayneutral ground for models to competeon. Last stated 5 days ago 3 Sept 2026 DR Dax Raad — holds since 2026-09-03 — tap for who they are Same subject: A machine can complete the task and completely miss the job -- coding makes this easy to see precisely because its tasks are so legible and verifiable. — tap to centre the map on it A machine can complete the task andcompletely miss the job -- codingmakes this easy to see preciselybecause its tasks are so legible andverifiable. Last stated 4 weeks ago 14 Aug 2026 SP Sunil Pai — holds since 2026-08-14 — tap for who they are Same subject: AI agents and AI coding will run on servers and from the cloud first, not on your laptop. — tap to centre the map on it AI agents and AI coding will run onservers and from the cloud first,not on your laptop. Last stated 2 months ago 28 Jun 2026 PL Pieter Levels — holds since 2026-06-28 — tap for who they are Same subject: Anthropic pulling Claude subscriptions from third-party harnesses like OpenCode is a mistake: playing a single match in a game of many rounds. — tap to centre the map on it Anthropic pulling Claudesubscriptions from third-partyharnesses like OpenCode is amistake: playing a single match in agame of many rounds. Last stated 5 months ago 8 Apr 2026 DH David Heinemeier Hansson — holds since 2026-04-08 — tap for who they are Same subject: Coding agents should read the shared AGENTS.md convention; insisting on a tool-specific CLAUDE.md creates split-brain problems across a team. — tap to centre the map on it Coding agents should read the sharedAGENTS.md convention; insisting on atool-specific CLAUDE.md createssplit-brain problems across a team. Last stated 2 weeks ago 25 Aug 2026 TL Tobias Lütke — holds since 2026-08-25 — tap for who they are Same subject: A team that slows down and reads every pull request and every line of code should expect only a 30 to 50 percent productivity lift from AI. — tap to centre the map on it A team that slows down and readsevery pull request and every line ofcode should expect only a 30 to 50percent productivity lift from AI. Last stated 2 months ago 15 Jul 2026 DH Dex Horthy — holds since 2026-07-15 — tap for who they are Same subject: Having an AI agent write tests costs tokens in the short run and saves them in the long run, because the classic failure is that the code does not work. — tap to centre the map on it Having an AI agent write tests coststokens in the short run and savesthem in the long run, because theclassic failure is that the codedoes not work. Last stated 2 months ago 1 Jul 2026 KB Kent Beck — holds since 2026-07-01 — tap for who they are Same subject: A language-agnostic conformance suite is the most powerful thing you can hand a coding agent, because the whole instruction becomes: write code until these tests pass. — tap to centre the map on it A language-agnostic conformancesuite is the most powerful thing youcan hand a coding agent, because thewhole instruction becomes: writecode until these tests pass. Last stated 6 months ago 19 Mar 2026 SW Simon Willison — holds since 2026-03-19 — tap for who they are
same subject or similar wordinga cloud: claims about one subject, named for itbar: when it was last stated, on a scale from 2015 to today — full is todaya face: someone on record holding the claim — tap it for who they are

At the centre Codex team's new features likely do not work well with models not trained on that mechanism Last stated 5 Sept 2026 · 3 days ago Holds Mario Zechner Read this korrent →