Critic Argues Google Fails To Train Models, Rendering Compute Advantage Moot
Data restrictions prevent Google from building a training flywheel.
Entities: Google, Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
In a reply on X, @teortaxesTex argued that Google's AI problem "goes far beyond" distillation and that "they also don't train, so the compute point is moot," weighing in on a broader debate over whether Google's hardware position should give it an advantage. The claim is still just that: a pointed opinion from one account, not an independently verified finding. A second reply from @scaling01 backed the gist with a shorter read, calling it a "Google skill issue."
Combined views
6.2K
3 posts, first seen 6h ago
Critic Argues Google Fails To Train Models, Rendering Compute Advantage Moot
Data restrictions prevent Google from building a training flywheel.
Entities: Google, Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)
In a reply on X, @teortaxesTex argued that Google's AI problem "goes far beyond" distillation and that "they also don't train, so the compute point is moot," weighing in on a broader debate over whether Google's hardware position should give it an advantage. The claim is still just that: a pointed opinion from one account, not an independently verified finding. A second reply from backed the gist with a shorter read, calling it a "Google skill issue."
Reactions from ranked influencers
3 posts
@scaling01 Reasonable prior but Google's problem goes far beyond that they do not distill they also don't train, so the compute point is moot
@teortaxesTex there's definitely a Google skill issue at play here
@teortaxesTex @scaling01 they are also not allowed to train on any kind of customer data (biggest problem for them, they don't have a data flywheel)