Tap a claim on the ring to put it at the centre.
← It is fundamentally difficult to design a reward function that…
17 connected korrents · 17 moments from 1 Mar 1999 to 17 Oct 2026. Nearly all of them are about reinforcement learning .
Everything filed under AI alignment
AI alignment
Everything filed under design
design
Everything filed under LLMs
LLMs
Everything filed under neural networks
neural networks
Everything filed under benchmarks
benchmarks
Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject Same subject
Read this korrent: It is fundamentally difficult to design a reward function that accurately captures the intended goal in reinforcement learning.
It is fundamentally difficult to design a reward function that accurately captures the intended goal in reinforcement learning.
Last stated 2 years ago
28 Nov 2024
LW
Lilian Weng — holds since 2024-11-28 — 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 a reasoning model unless it has been trained with reinforcement learning with 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: A model trained with reinforcement learning might discover a language of its own that is better for thinking than English. — tap to centre the map on it
A model trained with reinforcement learning might discover a language of its own that is better for thinking than English.
Last stated 2 years ago
5 Feb 2025
AK
Andrej Karpathy — holds since 2025-02-05 — tap for who they are
Same subject: A verifiable task can be optimised by reinforcement learning until a neural network performs it extremely well. — tap to centre the map on it
A verifiable task can be optimised by reinforcement learning until a neural network performs it extremely well.
Last stated a year ago
17 Nov 2025
AK
Andrej Karpathy — holds since 2025-11-17 — tap for who they are
Same subject: Anything precise enough to benchmark is also precise enough to optimize for, and modern RL training increasingly resembles the shape of the benchmarks themselves. — tap to centre the map on it
Anything precise enough to benchmark is also precise enough to optimize for, and modern RL training increasingly resembles the shape of the benchmarks themselves.
Last stated 7 months ago
16 Apr 2026
SK
Sayash Kapoor — holds since 2026-04-16 — tap for who they are
Same subject: Before training a base model with RL to emit think tokens, it is better to do some continued pre-training or supervised fine-tuning with those tokens present. — tap to centre the map on it
Before training a base model with RL to emit think tokens, it is better to do some continued pre-training or supervised fine-tuning with those tokens present.
Last stated a month ago
10 Oct 2026
SR
Sebastian Raschka — holds since 2026-10-10 — tap for who they are
Same subject: Capability does not generalise for free: a model that will move mountains on an agentic task still tells the same bad joke it told five years ago. — tap to centre the map on it
Capability does not generalise for free: a model that will move mountains on an agentic task still tells the same bad joke it told five years ago.
Last stated 8 months ago
20 Mar 2026
AK
Andrej Karpathy — holds since 2026-03-20 — tap for who they are
Same subject: Current models are unsafe because they are not smart enough, not because they are too smart. — tap to centre the map on it
Current models are unsafe because they are not smart enough, not because they are too smart.
Last stated 2 months ago
13 Sept 2026
FC
François Chollet — holds since 2026-09-13 — tap for who they are
Same subject: Distillation is simpler than reinforcement learning, but reinforcement learning is the state of the art and distillation a more efficient approximation. — tap to centre the map on it
Distillation is simpler than reinforcement learning, but reinforcement learning is the state of the art and distillation a more efficient approximation.
Last stated a month ago
17 Oct 2026
SR
Sebastian Raschka — holds since 2026-10-17 — tap for who they are
Same subject: A company that uses an algorithm to nudge customers into behavior that isn't good for them is doing something bad. — tap to centre the map on it
A company that uses an algorithm to nudge customers into behavior that isn't good for them is doing something bad.
Last stated a year ago
7 Oct 2025
MM
Mr. Money Mustache — holds since 2025-10-07 — tap for who they are
Same subject: AI alignment desperately needs good empirical tests that distinguish between the ways things could go, designed so they are not gamed or over-interpreted. — tap to centre the map on it
AI alignment desperately needs good empirical tests that distinguish between the ways things could go, designed so they are not gamed or over-interpreted.
Last stated 4 years ago
12 May 2023
AC
Ajeya Cotra — holds since 2023-05-12 — tap for who they are
Same subject: AI chatbots are tuned to produce an answer that feels satisfying to the user rather than one that is correct. — tap to centre the map on it
AI chatbots are tuned to produce an answer that feels satisfying to the user rather than one that is correct.
Last stated 2 months ago
15 Sept 2026
ML
Michael Lopp — holds since 2026-09-15 — tap for who they are
Same subject: AI works best when it solves a specific problem for a particular community rather than trying to be all things to all people. — tap to centre the map on it
AI works best when it solves a specific problem for a particular community rather than trying to be all things to all people.
Last stated 7 months ago
28 Apr 2026
AD
Anil Dash — holds since 2026-04-28 — tap for who they are
Same subject: Aligning multi-agent AI systems is fundamentally a problem of institutional and political design, not model training. — tap to centre the map on it
Aligning multi-agent AI systems is fundamentally a problem of institutional and political design, not model training.
Last stated 3 months ago
21 Aug 2026
RK
Rohit Krishnan — holds since 2026-08-21 — tap for who they are
Same subject: Computers and software should always serve the person directly using them, not the interests of third parties. — tap to centre the map on it
Computers and software should always serve the person directly using them, not the interests of third parties.
Last stated 4 months ago
31 Jul 2026
AK
Andrew Kelley — holds since 2026-07-31 — tap for who they are
Same subject: A browser works better as a single window than as multiple separate windows. — tap to centre the map on it
A browser works better as a single window than as multiple separate windows.
Last stated 2 months ago
15 Sept 2026
CC
Chris Coyier — holds since 2026-09-15 — tap for who they are
Same subject: A design has to be kept simple for reliability and intellectual control, but nobody yet knows how to reach simplicity systematically. — tap to centre the map on it
A design has to be kept simple for reliability and intellectual control, but nobody yet knows how to reach simplicity systematically.
Last stated 28 years ago
1 Mar 1999
ED
Edsger W. Dijkstra — holds since 1999-03-01 — tap for who they are
Same subject: A designer's edge now is pulling things back, making them simpler and more understandable, because everyone can kind of execute right now. — tap to centre the map on it
A designer's edge now is pulling things back, making them simpler and more understandable, because everyone can kind of execute right now.
Last stated 4 months ago
29 Jul 2026
CD
Charlie Deets — holds since 2026-07-29 — tap for who they are
same subject or similar wording a cloud: claims about one subject, named for it bar: when it was last stated, on a scale from 1999 to today (stretched back to the oldest claim here) — full is today a face: someone who holds the claim — tap it for who they are
At the centre
It is fundamentally difficult to design a reward function that accurately captures the intended goal in reinforcement learning.
Last stated 28 Nov 2024 · 2 years ago
Holds Lilian Weng
Read this korrent →
Same subject: reinforcement learning
A model should not be called a reasoning model unless it has been trained with reinforcement learning with verifiable rewards.
Last stated 3 Oct 2026 · 2 months ago
Holds Sebastian Raschka
Same subject: reinforcement learning
A model trained with reinforcement learning might discover a language of its own that is better for thinking than English.
Last stated 5 Feb 2025 · 2 years ago
Holds Andrej Karpathy
Same subject: reinforcement learning
A verifiable task can be optimised by reinforcement learning until a neural network performs it extremely well.
Last stated 17 Nov 2025 · a year ago
Holds Andrej Karpathy
Same subject: reinforcement learning
Anything precise enough to benchmark is also precise enough to optimize for, and modern RL training increasingly resembles the shape of the benchmarks themselves.
Last stated 16 Apr 2026 · 7 months ago
Holds Sayash Kapoor
Same subject: reinforcement learning
Before training a base model with RL to emit think tokens, it is better to do some continued pre-training or supervised fine-tuning with those tokens present.
Last stated 10 Oct 2026 · a month ago
Holds Sebastian Raschka
Same subject: reinforcement learning
Capability does not generalise for free: a model that will move mountains on an agentic task still tells the same bad joke it told five years ago.
Last stated 20 Mar 2026 · 8 months ago
Holds Andrej Karpathy
Same subject: reinforcement learning
Current models are unsafe because they are not smart enough, not because they are too smart.
Last stated 13 Sept 2026 · 2 months ago
Holds François Chollet
Same subject: reinforcement learning
Distillation is simpler than reinforcement learning, but reinforcement learning is the state of the art and distillation a more efficient approximation.
Last stated 17 Oct 2026 · a month ago
Holds Sebastian Raschka
Same subject: AI alignment
A company that uses an algorithm to nudge customers into behavior that isn't good for them is doing something bad.
Last stated 7 Oct 2025 · a year ago
Holds Mr. Money Mustache
Same subject: AI alignment
AI alignment desperately needs good empirical tests that distinguish between the ways things could go, designed so they are not gamed or over-interpreted.
Last stated 12 May 2023 · 4 years ago
Holds Ajeya Cotra
Same subject: AI alignment
AI chatbots are tuned to produce an answer that feels satisfying to the user rather than one that is correct.
Last stated 15 Sept 2026 · 2 months ago
Holds Michael Lopp
Same subject: design
AI works best when it solves a specific problem for a particular community rather than trying to be all things to all people.
Last stated 28 Apr 2026 · 7 months ago
Holds Anil Dash
Same subject: design
Aligning multi-agent AI systems is fundamentally a problem of institutional and political design, not model training.
Last stated 21 Aug 2026 · 3 months ago
Holds Rohit Krishnan
Same subject: design
Computers and software should always serve the person directly using them, not the interests of third parties.
Last stated 31 Jul 2026 · 4 months ago
Holds Andrew Kelley
Same subject: simplicity
A browser works better as a single window than as multiple separate windows.
Last stated 15 Sept 2026 · 2 months ago
Holds Chris Coyier
Same subject: simplicity
A design has to be kept simple for reliability and intellectual control, but nobody yet knows how to reach simplicity systematically.
Last stated 1 Mar 1999 · 28 years ago
Holds Edsger W. Dijkstra
Same subject: simplicity
A designer's edge now is pulling things back, making them simpler and more understandable, because everyone can kind of execute right now.
Last stated 29 Jul 2026 · 4 months ago
Holds Charlie Deets