Tap a claim on the ring to put it at the centre.
← Spending compute at inference time to search over plausible solutions…
8 connected korrents · 5 moments on record from 19 Jun 2024 to 31 Jul 2026.
Drag to move · pinch or scroll to zoom
same subject or similar wording
At the centre
Spending compute at inference time to search over plausible solutions is the general fix for unreliable long agent runs.
Holds JDJeff Dean
Read this korrent →
-
Similar wording
Inference, not training, is now the constraint, and specialized low-energy hardware will beat general-purpose GPUs and TPUs on latency.
Holds JDJeff Dean
-
Similar wording
Real reasoning will have arrived when spending more compute at inference time reliably buys a dramatically better answer.
Holds ASAravind Srinivas
-
Similar wording
Long-running agents fail because they drift off the distribution they were trained on, and degrade further the farther out they get.
Holds JDJeff Dean
-
Similar wording
Agents can already be run for days or weeks on a single hard problem, and almost nobody has internalised that.
Holds JDJeff Dean
-
Similar wording
Demand for inference may be growing exponentially while GPU production can only grow linearly, and the tightening happens where those two lines cross.
Holds
Dax Raad
-
Similar wording
Orchestrating thousands of agents is a new form of test-time compute, a fourth lever alongside network size, training data and training flops.
Holds
Boris Cherny
-
Similar wording
Inference is a high-margin business even for a middleman: renting GPUs at scale, some models carry an eighty per cent margin over their sticker price.
Holds
Dax Raad
-
Similar wording
Knowing something yourself will stay faster than asking a model for it, because a lookup in your own head beats a round trip to an agent.
Holds
Patrick Collison