What does an equal odds game look like for a GM vs an engine?

Question

This question began an interesting discussion on handicaps for engines that play against GMs:

Engine-invented variations/hypotheticals

This led me to wonder, are there differences in handicaps in terms of how we measure advantage?

For instance would a material advantage be preferable to space or tempo advantages?

Not sure what the success metric would look like, but i'd like to see the mean permutation on what degree of material advantage places the odds 50/50 for engine vs GM. Then compare that board layout to what tempo advantage would be needed to place the odds at 50/50.

Perhaps if we don't have engines that emulate GMs, we can just use lichess records?

Answer 1

As of right now, the "equal odds" between a GM and an engine is about two pawns. I believe Hikaru Nakamura made that comment at some point in this match (I did not watch the entire video). For another perspective, this is what happened when GM David Smerdon played against Komodo with knight odds (GM Smerdon won 5-1).

Note that two pawn odds (White starts without b- and c- pawns) gives a starting eval of about -3.0, while knight odds (White begins without the b1-knight) gives a starting eval of about -5.0. Comparatively 4-tempi odds (White plays e4, d4, Nf3 and Nc3 at start of the game, before Black gets to move) is only about +2.

Do note that giving White five "free" tempi leads to a forced win. For five non-free tempi, you'd have to analyze the specific position.