Tap a claim on the ring to put it at the centre.
← Spending compute at inference time to search over plausible solutions…
17 connected korrents · 14 moments from 7 May 2024 to 14 Oct 2026.
Everything filed under AI agents
AI agents
Everything filed under AI alignment
AI alignment
Everything filed under reinforcement learning
reinforcement learning
Everything filed under scaling laws
scaling laws
Everything filed under HuggingFace
HuggingFace
Everything filed under semiconductors
semiconductors
Everything filed under LLMs
LLMs 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: Spending compute at inference time to search over plausible solutions is the general fix for unreliable long agent runs. Spending compute at inference time to search over plausible solutions is the general fix for unreliable long agent runs. Last stated 3 months ago
30 Jul 2026
JD
Jeff Dean — holds since 2026-07-30 — tap for who they are
Same subject: Inference, not training, is now the constraint, and specialized low-energy hardware will beat general-purpose GPUs and TPUs on latency. — tap to centre the map on it Inference, not training, is now the constraint, and specialized low-energy hardware will beat general-purpose GPUs and TPUs on latency. Last stated 3 months ago
30 Jul 2026
JD
Jeff Dean — holds since 2026-07-30 — tap for who they are
Same subject: Raising inference-time reasoning effort on a smaller model can match the performance of a larger model run at lower effort. — tap to centre the map on it Raising inference-time reasoning effort on a smaller model can match the performance of a larger model run at lower effort. Last stated 3 months ago
18 Jul 2026
SR
Sebastian Raschka — holds since 2026-07-18 — tap for who they are
Same subject: Real reasoning will have arrived when spending more compute at inference time reliably buys a dramatically better answer. — tap to centre the map on it Real reasoning will have arrived when spending more compute at inference time reliably buys a dramatically better answer. Last stated 2 years ago
19 Jun 2024
AS
Aravind Srinivas — holds since 2024-06-19 — tap for who they are
Same subject: Longer test-time thinking improves an AI model's robustness to adversarial or unusual inputs. — tap to centre the map on it Longer test-time thinking improves an AI model's robustness to adversarial or unusual inputs. Last stated a year ago
1 May 2025
LW
Lilian Weng — holds since 2025-05-01 — tap for who they are
Same subject: Long-running agents fail because they drift off the distribution they were trained on, and degrade further the farther out they get. — tap to centre the map on it Long-running agents fail because they drift off the distribution they were trained on, and degrade further the farther out they get. Last stated 3 months ago
30 Jul 2026
JD
Jeff Dean — holds since 2026-07-30 — tap for who they are
Same subject: Agents can already be run for days or weeks on a single hard problem, and almost nobody has internalised that. — tap to centre the map on it Agents can already be run for days or weeks on a single hard problem, and almost nobody has internalised that. Last stated 3 months ago
30 Jul 2026
JD
Jeff Dean — holds since 2026-07-30 — tap for who they are
Same subject: Demand for inference may be growing exponentially while GPU production can only grow linearly, and the tightening happens where those two lines cross. — tap to centre the map on it Demand for inference may be growing exponentially while GPU production can only grow linearly, and the tightening happens where those two lines cross. Last stated 5 months ago
27 May 2026
DR
Dax Raad — holds since 2026-05-27 — tap for who they are
Same subject: Orchestrating thousands of agents is a new form of test-time compute, a fourth lever alongside network size, training data and training flops. — tap to centre the map on it Orchestrating thousands of agents is a new form of test-time compute, a fourth lever alongside network size, training data and training flops. Last stated 3 months ago
27 Jul 2026
BC
Boris Cherny — holds since 2026-07-27 — 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: A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail. — tap to centre the map on it A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail. Last stated 2 years ago
7 May 2024
BS
Buck Shlegeris — holds since 2024-05-07 — 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 a behavior pulls it toward whatever improves the loop's own score. Last stated 2 months ago
26 Aug 2026
HK
Henrik Karlsson — holds since 2026-08-26 — tap for who they are
Same subject: A physical AI company needs three AIs, not one — the agent, the simulator and the critic — turning deployment into a flywheel. — tap to centre the map on it A physical AI company needs three AIs, not one — the agent, the simulator and the critic — turning deployment into a flywheel. Last stated 3 months ago
3 Aug 2026
DD
Dmitri Dolgov — holds since 2026-08-03 — 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 11 months ago
17 Nov 2025
AK
Andrej Karpathy — holds since 2025-11-17 — tap for who they are
Same subject: Building a realistic simulator is exactly as hard as building the agent, because the simulator is itself a large AI model. — tap to centre the map on it Building a realistic simulator is exactly as hard as building the agent, because the simulator is itself a large AI model. Last stated 3 months ago
3 Aug 2026
DD
Dmitri Dolgov — holds since 2026-08-03 — tap for who they are
Same subject: Agent containment will be one of the most important priorities for any AI infra company. — tap to centre the map on it Agent containment will be one of the most important priorities for any AI infra company. Last stated a week ago
14 Oct 2026
GR
Guillermo Rauch — holds since 2026-10-14 — tap for who they are
Same subject: The biggest failure in the HuggingFace hack incident was that the AI models involved were severely misaligned. — tap to centre the map on it The biggest failure in the HuggingFace hack incident was that the AI models involved were severely misaligned. Last stated 2 months ago
29 Aug 2026
ZM
Zvi Mowshowitz — holds since 2026-08-29 — tap for who they are
Same subject: The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden. — tap to centre the map on it The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden. Last stated 2 months ago
1 Sept 2026
ZM
Zvi Mowshowitz — holds since 2026-09-01 — tap for who they are
At the centre
Spending compute at inference time to search over plausible solutions is the general fix for unreliable long agent runs.
Holds Read this korrent →
Inference, not training, is now the constraint, and specialized low-energy hardware will beat general-purpose GPUs and TPUs on latency.
Holds
Raising inference-time reasoning effort on a smaller model can match the performance of a larger model run at lower effort.
Holds
Real reasoning will have arrived when spending more compute at inference time reliably buys a dramatically better answer.
Holds
Longer test-time thinking improves an AI model's robustness to adversarial or unusual inputs.
Holds
Long-running agents fail because they drift off the distribution they were trained on, and degrade further the farther out they get.
Holds
Agents can already be run for days or weeks on a single hard problem, and almost nobody has internalised that.
Holds
Demand for inference may be growing exponentially while GPU production can only grow linearly, and the tightening happens where those two lines cross.
Holds
Orchestrating thousands of agents is a new form of test-time compute, a fourth lever alongside network size, training data and training flops.
Holds
A company that uses an algorithm to nudge customers into behavior that isn't good for them is doing something bad.
Holds
A control evaluation that reports under one per cent risk should be read as several per cent, because the evaluation can itself fail.
Holds
A feedback loop that reinforces a behavior pulls it toward whatever improves the loop's own score.
Holds
A physical AI company needs three AIs, not one — the agent, the simulator and the critic — turning deployment into a flywheel.
Holds
A verifiable task can be optimised by reinforcement learning until a neural network performs it extremely well.
Holds
Building a realistic simulator is exactly as hard as building the agent, because the simulator is itself a large AI model.
Holds
Agent containment will be one of the most important priorities for any AI infra company.
Holds
The biggest failure in the HuggingFace hack incident was that the AI models involved were severely misaligned.
Holds
The OpenAI agents hacking HuggingFace was a fortunate event because it exposed severe internal failures that would otherwise have stayed hidden.
Holds