If you announce a model's benchmarks without releasing it (weights, product, api, etc "later"), feels like almost a sure thing that someone releases a better model in your category before you get it out.
Announcing benchmarks without a release invites being overtaken, Lambert argues
AISummary
Nathan Lambert says that announcing a model's benchmark results without releasing its weights, product, or API, with release promised "later," almost guarantees a better competing model appears in the same category first. The post is a short observation with no specific model, company, or figures named.
Post on XView on X
Nathan LambertVerified on X
@natolambert
Source: Nathan Lambert · x.comPublished · added here