AI & Tech
Computer Use Progress Slower Because AI Cannot Grind Against Real Websites
Dwarkesh Patel Podcast
The next big breakthrough will be AIs learning on the job
"You can't just have 1,000 agents go try the same checkout flow on Amazon to get better at using websites because Andy Jassy will find your bots and shut your ass down. You can solve this by making clones of Slack and Gmail and all the other common applications and websites. But at least currently, this is a very labor-intensive and unscalable way to build environments."
The speaker reveals that progress on AI computer use capabilities lags behind other domains because training requires replayable simulators, and companies like Amazon will block bot training on real websites. This forces labs to build labor-intensive clones of applications, highlighting an underrated bottleneck in AI development that won't be solved until AIs can build high-fidelity application clones themselves.
From this episode
Dwarkesh Patel Podcast