Bespoke compliment generator… I’m using that

It's still a completely dubious chart, maybe more so given how vague it is. It doesn't specify which tasks, which humans, which models etc.

The chart is confusing, but the Y axis is "how much time would it take a human to do a task that AI can now do 50% of the time" so the time part refers to human time spent on the task.

Ed, one small mistake in your post, re: the a16z chart about model task completion. You wrote: "it’s measuring how long a model can take to do something[...] Just because a model can run for hours doesn’t mean the outputs will be good"

I'm able to see it still: www.justice.gov/epstein/file...

Well written, thanks Reece

What’s the cause of the missing data