AI Sensei

How is AI Sensei different than OGS AI analysis? Watched a youtube video and found some convenient features, but did not feel much needed cause I have OGS AI analysis already.

But any insights are appreciated.

1 Like

I think you can go more in depth on different variations, kind of like you’d have if you ran the AI locally. OGS only does the top one or two lines.

But I only use OGS AI and think it gets me what I need most of the time. There are times when I wish I had more breadthwise room to explore in case the AI is ignoring local L&D though.

EDIT: I don’t think you can explore in AI sensei any more than you can in OGS. That said, you do seem to get more bang for your buck (only looking at AI features, OGS is more featureful overall of course)

1 Like

OGS

Playouts Approximate strength Price
Supporter 125 Amateur Dan level player <3€/month?
Kyu Supporter 300 Professional level player 3€/month
Dan Supporter 800 Strong professional level player 5€/month
Pro Supporter 2000 Beyond most professional level players 10€/month

Aisensei

Plan Playouts Variation Length Sample Monthly Price
Free 50 about 4 moves Link FREE
Basic 500 about 10 moves Link 5.95 $ or 4.95 €
Dan 2500 about 15 moves Link 11.95 $ or 9.95 €
Pro 10000 about 18 moves Link 27.95 $ or 24.95 €

I think they’re very similar. I’m not 100% on the OGS supporter level and plans because there’s also month/year donations and one off donations as opposed to recurring ones.

I guess a key difference is the free 50 move playouts from aisensei, whereas the free OGS is the after game win rate and top3 moves.

Someone more in the know can maybe compare on what playouts means, whether the numbers are comparable or better etc for one given the length of the sequences etc (I assume a playout is the ai finishing the game from that point and evaluating the position?). I mean to me, they look comparable with the same prices.

I think they’re more or less the similar in what’s offered with AI, both have Katago and LeelaZero, winrate and score estimation (from katago) and graphs. Ai sensei has a feature to filter by points loss however, so it makes it easier to jump to or filter by moves which are mistakes of over a fixed size (more than 5 points etc). Ai sensei can also show territory prediction on the board.

Ai sensei also has a feature in beta for making puzzles from games which I haven’t tried out recently since it’s been updated to customize the solutions. Before it was just mainly a “guess the ai move” sort of puzzle.

Anyway you can easily test out both the Ai sensei and OGS features.

Ai sensei make an account or check out some of their sample reviews eg https://ai-sensei.com/game/ZuAQOdZvFjbS98TtZ3zomIyjBfi2/-M-z3WcUA0zos6xAhkiR

OGS (you already have it but for others) just find some games with different levels of site support, eg a recent enough Euro Pro game Play Go at online-go.com! | OGS

2 Likes

My AI-Sensei workflow is mostly this:

  • upload every game, which creates a complete analysis of the game in one go
  • review, and make every position where I want to remember the (arguably) correct move into a problem (that’s a single click mostly)
  • keep up with the spaced repetition of all these problems

There are some caveats, mainly that I have only Basic plan (500 playouts), so the AI may miss something or even completely misjudge a situation. I therefore only make problems where I can rationalize that the move I set as a solution actually makes sense to me. The trial board is available anytime to explore variations manually.

What’s still missing, but apparently in the works, is a better catalogue of games and problems, and »live« analysis of ad hoc variations. What I can see is that the developers are very receptive to feedback (using the very prominent feedback button), and there is constant and careful improvement. I think that the direction is really something that you can view as a 先生.

4 Likes

Based on the table above giving pro strength as 300 playouts and strong pro as 800 playouts for the OGS AIs I guess that 500 playouts should be alright for almost all situations as an amateur player - if it is not about reading out complicated capturing races? :slight_smile:

1 Like

Probably still a good idea to stick to moves you can justify, though.

3 Likes

That the AI in this setting is somehow »pro strength« doesn’t mean that each and every move it proposes is strictly better than anything that e. g. I can come up with. There still seem to be horizon effects in some cases. For example, I see the AI propose to leave some isolated weak stones alone and take a big point outside, but when I see the proposed continuation, I get the impression that it simply could not imagine the stones to die completely and still assigned some value to them. I hope that I can analyse such things better when live mode is implemented.

2 Likes

AI has crazy good direction, even without playouts, inhumane life and death skills and is super flexible.

I frequently use it to go over my games too, and many times, especially if the variations are short, I have no clue what it is aiming at lmao. I guess the live mode on AI Sensei could make these cases clearer. ^^

2 Likes

Altho not necessarily more playable. : P

I have OGS site supporter AI analysis and free AI Sensei account. I do like the AI sensei feature of jumping to mistakes, i.e. moves that lost more than a certain number of points or win percentage that you set. Also, you upload SGF and it analyses instantly. But then I like that the OGS one shows the points / win rate change on each intersection that AI considered (on AI Sensei you have to click the variation and it only the shows the points / win rate change below the board). And OGS one runs straight after a game automatically. And the 300 playouts OGS Vs 50 playouts AI sensei does seem to make a difference in terms of longer variations. So, overall, I prefer OGS and just have AI sensei as a backup, e.g. if OGS upload of SGFs isn’t working (as was the case the other day).

Incidentally, have you seen the analysis in chess.com? It gives you stats on your game like 10 moves were best move, 5 were good, 3 were inaccurate, 1 mistake, 1 blunder, etc. Presumably these are just chess AI win rate thresholds. Might be a nice addition to Go AI analysis (on OGS to maintain the edge on the competition :wink:)

3 Likes

Thanks everyone. It sounds like OGS AI is good enough for me. I have a two day free trial voucher, thus am curious.

2 Likes

Yeah, OGS AI is a very nice tool, no need to quit and download the game, it analyses straight after having played. And it is a good way to support the server too. :smiley:

2 Likes

It’s worth noting that ZBaduk can transfer games from OGS straight to your ZBaduk account.

Prices are more competitive in my opinion
and there aren’t any calculation limits (other than time).

Spoiler: I’m the author.