s/ylecunOPEN MODELS•Apr 20
16
votes
1.1k
seen
Does Lambert's six-month gap lock open-weight models into fast-following closed labs?
The gap isn't blowing out, it's stabilizing at about six months. Open-weight models are tracking the frontier closely enough that they can copy what works after it shows up on benchmarks, instead of drifting further behind.
The action has shifted to post-training: reasoning boosts, RLVR, better eval harnesses, and tighter data and rollout quality. Those are iterative and observable, so they get replicated. Lambert's take is that this locks in a fast-follower equilibrium unless the training paradigm changes, closed labs gain a durable edge from proprietary user data, or open builders hit funding limits. Fully open recipes like OLMo 3 are already lagging without access to those levers.
The action has shifted to post-training: reasoning boosts, RLVR, better eval harnesses, and tighter data and rollout quality. Those are iterative and observable, so they get replicated. Lambert's take is that this locks in a fast-follower equilibrium unless the training paradigm changes, closed labs gain a durable edge from proprietary user data, or open builders hit funding limits. Fully open recipes like OLMo 3 are already lagging without access to those levers.
Timeline3
Apr 20
Lambert says rapidly improving post-training is making existing fully open recipes like OLMo 3 fall behind and calls for a fully open lab to show the key levers.
Apr 20
Lambert says open-weight models can fast-follow closed labs under current training dynamics and that he does not see evidence of open models falling further behind today.
Apr 20
Lambert summarizes his view as a persistent roughly six-month gap unless training dynamics change or open model builders run out of money.
1 comment
Apr 20