korrents

On the map

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

← Fine-tuned model organisms are weak test cases, because fine-tuning…

17 connected korrents · 15 moments from 8 Aug 2023 to 5 Nov 2026.

Everything filed under reinforcement learning reinforcement learning Everything filed under LLMs LLMs Everything filed under LLMs LLMs Everything filed under taste taste Everything filed under design design Same subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subjectSame subject Read this korrent: Fine-tuned model organisms are weak test cases, because fine-tuning gives them no reason not to think about the objective all the time. Fine-tuned model organisms are weak testcases, because fine-tuning gives them noreason not to think about the objectiveall the time. Last stated 11 months ago 20 Dec 2025 NN Neel Nanda — holds since 2025-12-20 — tap for who they are Same subject: A model you have to fine-tune for each thing you want it to do is not a general-purpose model. — tap to centre the map on it A model you have to fine-tune foreach thing you want it to do is nota general-purpose model. Last stated 3 months ago 12 Aug 2026 CF Chelsea Finn — holds since 2026-08-12 — tap for who they are Same subject: Fine-tuning is probably not dead, but companies should not rush into it because many underestimate the data it requires. — tap to centre the map on it Fine-tuning is probably not dead,but companies should not rush intoit because many underestimate thedata it requires. Last stated 2 weeks ago 5 Nov 2026 AR Aishwarya Naresh Reganti — holds since 2026-11-05 — tap for who they are Same subject: A large context window does not replace fine-tuning, because fine-tuning serves the separate purpose of instilling preferences and demonstrating a task. — tap to centre the map on it A large context window does notreplace fine-tuning, becausefine-tuning serves the separatepurpose of instilling preferencesand demonstrating a task. Last stated 3 years ago 22 Mar 2024 EG Elad Gil — holds since 2024-03-22 — tap for who they are Same subject: Fine-tuning-based alignment does not remove a model's worrying knowledge and abilities; it only teaches the model not to output them. — tap to centre the map on it Fine-tuning-based alignment does notremove a model's worrying knowledgeand abilities; it only teaches themodel not to output them. Last stated 3 years ago 8 Aug 2023 DA Dario Amodei — holds since 2023-08-08 — tap for who they are Same subject: Models look far better on evals than they are in the world because researchers, inadvertently, take inspiration from the evals when they build RL environments. — tap to centre the map on it Models look far better on evals thanthey are in the world becauseresearchers, inadvertently, takeinspiration from the evals when theybuild RL environments. Last stated 12 months ago 25 Nov 2025 IS Ilya Sutskever — holds since 2025-11-25 — 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 5 months ago 26 Jun 2026 NB Noam Brown — holds since 2026-06-26 — tap for who they are Same subject: Experimenting with mental models does not require high craft and is better done at speed, in the open, in iteration than by polishing something that is the wrong shape. — tap to centre the map on it Experimenting with mental modelsdoes not require high craft and isbetter done at speed, in the open,in iteration than by polishingsomething that is the wrong shape. Last stated 5 months ago 8 Jul 2026 MC Meaghan Choi — holds since 2026-07-08 — tap for who they are Same subject: Fine-tuning is hard to justify for a few thousand API calls but can make sense for millions of repetitive requests. — tap to centre the map on it Fine-tuning is hard to justify for afew thousand API calls but can makesense for millions of repetitiverequests. Last stated 2 weeks ago 5 Nov 2026 AR Aishwarya Naresh Reganti — holds since 2026-11-05 — tap for who they are Same subject: A feedback loop that reinforces a behavior pulls it toward whatever improves the loop's own score. — tap to centre the map on it A feedback loop that reinforces abehavior pulls it toward whateverimproves the loop's own score. Last stated 3 months ago 26 Aug 2026 HK Henrik Karlsson — holds since 2026-08-26 — tap for who they are Same subject: A format reward should only be added when the answer is already correct. — tap to centre the map on it A format reward should only be addedwhen the answer is already correct. Last stated a month ago 10 Oct 2026 SR Sebastian Raschka — holds since 2026-10-10 — tap for who they are Same subject: A model should not be called a reasoning model unless it has been trained with reinforcement learning with verifiable rewards. — tap to centre the map on it A model should not be called areasoning model unless it has beentrained with reinforcement learningwith verifiable rewards. Last stated 2 months ago 3 Oct 2026 SR Sebastian Raschka — holds since 2026-10-03 — tap for who they are Same subject: One pre-trained robot model now matches or beats the specialists that were fine-tuned with reinforcement learning for the very tasks they were built for. — tap to centre the map on it One pre-trained robot model nowmatches or beats the specialiststhat were fine-tuned withreinforcement learning for the verytasks they were built for. Last stated 3 months ago 12 Aug 2026 CF Chelsea Finn — holds since 2026-08-12 — tap for who they are Same subject: Reading RL transcripts after a post-training run to find problems in training environments seems a good general practice. — tap to centre the map on it Reading RL transcripts after apost-training run to find problemsin training environments seems agood general practice. Last stated 11 months ago 27 Dec 2025 NN Neel Nanda — holds since 2025-12-27 — tap for who they are Same subject: RL environments exist to make a model generalise, not to teach it each skill one at a time — exactly as pre-training does. — tap to centre the map on it RL environments exist to make amodel generalise, not to teach iteach skill one at a time — exactlyas pre-training does. Last stated 9 months ago 13 Feb 2026 DA Dario Amodei — holds since 2026-02-13 — tap for who they are Same subject: Behaviours like massively deluding humans or trying to seize control of assets could in principle generalise from reward hacking. — tap to centre the map on it Behaviours like massively deludinghumans or trying to seize control ofassets could in principle generalisefrom reward hacking. Last stated a year ago 8 Jul 2025 RG Ryan Greenblatt — holds since 2025-07-08 — tap for who they are Same subject: How aligned current models are depends a lot on things like their degree of reward hacking. — tap to centre the map on it How aligned current models aredepends a lot on things like theirdegree of reward hacking. Last stated 11 months ago 27 Dec 2025 NN Neel Nanda — holds since 2025-12-27 — tap for who they are Same subject: If whole categories of reward hack are undetectable by humans, then training against the hacks we do catch teaches models to cheat only where we cannot see. — tap to centre the map on it If whole categories of reward hackare undetectable by humans, thentraining against the hacks we docatch teaches models to cheat onlywhere we cannot see. Last stated 3 months ago 11 Aug 2026 RG Ryan Greenblatt — holds since 2026-08-11 — 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 who holds the claim — tap it for who they are

At the centre Fine-tuned model organisms are weak test cases, because fine-tuning gives them no reason not to think about the objective all the time. Last stated 20 Dec 2025 · 11 months ago Holds Neel Nanda Read this korrent →