the reason every model is bad at multimodal is because literally nobody except @vikhyatk is even trying there’s prob a lot of easy wins on the CUA path still to be found
Tzafon
TzafonFeb 27, 03:20
We showed model colored squares for a few hours. It learned to use a computer better than models trained on thousands of real screenshots.
@ainativefirm @vikhyatk pre-prometheus, @sherjilozair also did some excellent CUA work which didn't get nearly enough attention
@ainativefirm @vikhyatk both anthropic + perplexity correctly have identified that the next wave of AI power uses beyond coding is finance
114