AI review strength increase

is that new bug?
b+34 points on OGS review, but on OGS estimator and on free AI sensei b+ half


https://online-go.com/game/42536951

(after I do the only possible move, it suddenly realizes that its b+0.7, how could it not see this move before? :grinning: )

  1. Game is unfinished, i.e.

    • many borders open,
    • areas not yet settled

    so the Score Estimator estimates — and differently for every estimate:

    Screen Shot Safari Technology Preview Game Finished 2022-04-01 at 17.56.29 CET

    BTW I clicked “Full AI Review” for you … so now maybe different results.

  2. And we must keep in mind that it always assumes “best play” by both sides, plus it does NOT always assume the same “best play” with every new estimation, if I understand correctly.

34,6 looks like too big difference. Estimator gives it only if I pass

Maybe better use small net for not supporters, but more playouts than now?
These spikes make no sense.

go katago!

1 Like

In this game it appears the AI did not count blacks points in the top middle nor does it seem the AI counted whites points from the lower right on the board. dimitriesdream vs. jon.edwards

EDIT: Also, not sure if its do to the updates but during this game on the app for some reason the ACCEPT function wouldnt register my selection and me and my opponent had to “PASS” seversl times befor the game accepted our selections.

@dimitriesdream

Those are because the position was not settled:
N19 and R1 were still useful endgame plays.

I acknowledge your first screen capture seems weird, or like there might be a bug because below the blue “Full AI Review” button (and above the green “Back to Board” button) the estimate is “Black by 0.3” and inside the black-bar the estimate is “Black +34.6” ("Black by 34.6) - and additionally it is not showing the ten White stones (in the F12 area) as very alive… And that is interesting; but consider that the White stones are in atari and even though White can (easily) connect (by filling in the false eye at D12) and save those ten stones the six Black stones still have some aji < Aji at Sensei's Library > and thus Black can use them to make a living area/ shape in the corner… thus estimating the score is not easy/ depends-a-lot-on-next-moves… but the question still remains, why one estimate of 0.3 and one estimate of 34.6 BOTH in the first screen-capture/ at-the-same-time? Good question IMO! IDK, but maybe those two estimates were actually calculated at different times (even though they are displayed at the same time)? Or maybe two different algorithms (estimating methods) are being used?

Does anybody have any additional info?

…this is just my quick “first draft”, so probably not the best moves, but I believe it shows why I claim (at my OGS 3k level of understanding) there is some aji and how a living shape could be made in the upper left corner.

Sorry if I’m going off-topic here, but my point is “estimating” is… estimating, and not exact. Here are a few of my thoughts on how the game could have gone (kinda like “estimates” lol ) —> < 9k - 6k >