If all this is is an update on Deepseek, that's silly. The lesson isn't that people are going to use a little bit of compute to train extremely powerful models, it's that they are going to use a tremendous amount of compute to (unwisely) build tremendously powerful models.

Joe Weisenthal

@thestalwart · Twitter · 2025-01-27 10:51 UTC

Pre-market $META -5.3% $GOOGL -5.1% $NVDA -13% $ASML -9.7% $MSFT -6.7%