Dimakis doubts Claude will show superlinear gains over longer runs
@giffmana @MangQiuyang We’ll do it for longer Luke— it just costs monies 🤑. I’d be shocked if Claude code somehow picks up and becomes super linear though, do you think that can happen ?
@MangQiuyang I don't think you can just extrapolate the agents like this on a logx scale. If you were to cut off the human curve at 24h and extrapolate the same way, it would look a lot more flat.
No Digg Deeper questions have been answered for this story yet.
@AlexGDimakis @MangQiuyang Actually maybe it's not oddly spaced but the two ticks at 24 and 14 make it seem so.
Dunno but it looks like this happens for the humans? Claude does not really look more flat than humans in the area where you have measurements, only in the one you extrapolate.
I think the very oddly spaced x axis combined with perfectly linear extrapolation in visual (not data!) space does a lot of work here, i do not believe this extrapolation, while it is what carries the point.
Dunno but it looks like this happens for the humans? Claude does not really look more flat than humans in the area where you have measurements, only in the one you extrapolate.
I think the very oddly spaced x axis combined with perfectly linear extrapolation in visual (not data!) space does a lot of work here, i do not believe this extrapolation, while it is what carries the point.
@giffmana @MangQiuyang We’ll do it for longer Luke— it just costs monies 🤑. I’d be shocked if Claude code somehow picks up and becomes super linear though, do you think that can happen ?