AI Sensei has two features that are very good and perhaps would be something to implement on OGS.
The first is that for the AI review you can set the level of the student for the AI review so that it gives feedback that is easier to understand.
The second is that you can add problems from the review and use them for practice. They have a very good feature which is that you press add multiple problems and then the AI automatically adds the most serious misstakes from the kifu to your problem library so that you can practice.
Is it that the level is more like a filter where you don’t call a move a mistake if it’s over X points, where X is decided by your level.
If I’m a 7d maybe I might care about the 0.5-1 point mistakes, if I’m a 10kyu maybe I want to focus on the 6-10 point mistakes?
That kind of idea?
As a simpler system or maybe a first step toward something like that, maybe being able to copy game trees/single branch from one place to another, like a game to a demo board, or a demo board to a puzzle might be step one?
Just thinking from a development perspective, and if that functionality is there then you make a button that automates it in scenarios like an ai review etc
That “student level” setting … could use some work. For example, maybe it shows you a 30 point mistake. Then you look at the ai playout. Perhaps the followup consists of a couple of moves, and it’s a mistake the player could learn from. Often, though, the followup consists of 10-30 moves before one side benefits from the ai suggested move.
I am not sure how it works exactly, but I think it is connected to how many points loss a move must make to be marked as a misstake for the level you have given. I have it set to 6 kyu since I can solve tsumegos classified as 6-7 kyu on 101weiqi and I understand why the misstakes that are thus marked in the review as misstakes are misstakes. Now I usually use AI Sensei to review my games since I found it was more didactic than the AI review on OGS (although the AI review on OGS became a bit more helpful when I set it to not showing variations that included more than 7 moves). But I really like that I can just uppload the Kifu on AI Sensei and then press add multiple problems and than work with them in my problem library until I feel that I recognize and understand the position.
Whilst learning from others / shamlessly ripping off competitors does have its merits, I feel that OGS should use its limited development resources to improve the things OGS does that AI sensei does not, i.e playing games, rather than duplicating effort and becoming an AI sensei clone.
OGS predates AI, and the original design focused on correspondence games, and in terms communities. AI function is an add-on much much later and a sub-function of the review function. So indeed I’d imagine if we want to extend and make AI review an independent feature would require quite a bit of modifications.
But extending the existing features still might be doable. Especially if those are limited to paid more advanced AI reviews, then the effort might have a measurable payoff.
I wonder if AI-sensei’s feature is complete, why not just create an additional bottom that can automatically upload games to AI-sensei like Pandanet? Is there any agreement required for the integration? Or is there a worry that this might erode the OGS paid users?
It’s not automated that it’ll grab every game, but being able to just paste a url, or drag and drop an sgf is pretty good for saving clicks and time to be fair.
I used AISensei for a while, and my optimum would be:
In the AI Sensei settings, click on “Connect my OGS Account”, paste in the URL to your OGS profile (or an API token, or whatever; I’m not sure exactly what would make sense here), and click “Connect Account” (bonus points if multiple accounts can be connected)
Finish a game on OGS
Go to AI Sensei, to a page for this purpose
AI Sensei retrieves your recent games (but doesn’t waste time analyzing them, yet)
Find the game you just played, and click on “Analyze Game”
AI Sensei grabs that game and pulls it into their ecosystem, and things go as expected from there
This would definitely require AI-sensei’s team’s cooperation to change their UI and there is a connect to pandanet account on AI-sensei and looks like their only cooperation partner right now.
l have to say one thing about AI-sensei though is that its review function and discuss function aren’t connected, making the actual human reviews extremely difficult. I supposed it is not their focus (it’s called AI-sensei after all), but almost all the posts and discussions on AI-sensei are about why AI suggested some moves and why their moves are mistakes. And I got a feeling many users on AI-sensei don’t actually get much out of the AI reviews (especially SDK players). A lot of the “human replies” are just repeats of “generic answers” boil down to “AI said suggested moves are more urgent and your move can wait” (without even specify why)
Benjamin Teuber, the programmer of AI sensei (and also a top German player) has also worked on the pandanet newish official client so there is a connection there.
