2015 AlphaGo would still crush 2026 Fable in Go
OpenMined founder Andrew Trask claims the 2015 AlphaGo system would defeat a 2026 AI system in Go
The debate highlights single-task AI performance versus broader modern models
Positive users hail the AlphaGo vs Fable claim as a strong case for specialization beating generality in superintelligence, while negative users dismiss the matchup as trivial like a calculator outperforming a mathematician.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@iamtrask I wonder! Is there an open source engine now that beats original AlphaGo? If so I'd expect Fable to know that, realize it's the best way to win, and be able to download and run it
2015 AlphaGo would still crush 2026 Fable in Go

@iamtrask wouldn't a calculator crush fable in calculation speed?

@NathanpmYoung Another great example of the benefits of specialization over generality for achieving superintelligent performance.

Superintelligence is most "super" when it is specialized... not when it's general.
This is good for safety, concentration of power, privacy, etc... but it's bad if your goal is to build "the everything product" and go from a X Trillion dollar biz to an X+1 Trillion dollar biz.

@NathanpmYoung My point is that there's considerable evidence to suggest the outermost limits of performance require specialization more than generalization.

@iamtrask Don't you think 2026 Fable could train something that crushes AlphaGo

@iamtrask yeah, but my Texas Instruments calculator would crush Terence Tao at calculation. big whoop