As a lowly 5 kyu, I find AI a bit tricky to use.. I run into this problem where my opponent and I both look at a group and think "that's probably fine." But it's not fine, and the AI spends the next 100 or so moves swinging back and forth because whom-ever's turn it is is the obvious winner :D
Thanks for posting this, it's really great to have some advice for how to maximize the utility of the analysis tools that are available to us. Something as simple as having questions that you want an answer to makes a big difference in understanding and learning, vs. just looking at inflection points in the win rate graph. Also really appreciate the emphasis on ignoring the < 5 point mistakes when reviewing
This sounds solid. It is an extension of the principal that it is important to have a plan, because even if that plan fails at least you know what failed and how. There is a big point that is too obvious to say. Do not try to emulate the AI. You will never be able to do that. There, said it.
Nonsense. Plenty of AI innovations initially dismissed as incomprehensible are being studied by human players and incorporated intelligibly into human games even at the kyu level. I can give plenty of examples on request. A more accurate phrasing of your advice would be, "try to understand what you emulate (but also be aware that some lessons only come with practice)." If you get a piece of advice from an AI that resists all attempts to be understood, consider it provisional if you don't forget it altogether; if you get a piece of advice from an old Japanese fuseki book that resists all attempts to be understood (and as someone who's read plenty of Go books before AI became superhuman, trust me there is no shortage of under-explained, inscrutable, even inaccurate pro advice), consider it provisional if you don't forget it altogether. Blindly emulating humans is no better than blindly emulating AI (and if you do, at least the latter is objectively stronger, less prone to blind spots and overconfidence, and infinitely patient with you).
I created an account using your link to support you as usual especially as it's a paid promotion, but I was looking forward for basics :'( Guess I'll have to patiently wait for next week 😁
I wanted to stick with something anyone could access. As stated in the video, some people cant run katrain on their device. Also I liked the features of ai sensei, which also has a free option.
One typically doesn't mention the competition in a sponsored video, and as a free tool, I can't really sponsor much. Regardless, AI sensei is a great tool if you prefer something web-based, and if dwyrin found any features missing in particular in KaTrain I'd be happy to hear it.
By going back and taking the time to read out the variations you couldn't read out in game, as well as keeping basic principles in mind (make good shape, attack weak groups, play big moves) to see which of your moves were slow.
Well, usually during your games you'll have OH SHIT moments, where your opponent separates two of your groups, or kills a group you thought was alive, or something other that's huge for him. Going back to those and seeing what you could have done to avoid it is a good start. Other things to keep in mind: My opponent did X and I responded locally. Was X really sente for him? I got sente and did X. Should I have done something else?
I think people misunderstand how to use AI way too often they borderline misunderstand it so bad that they try to become the AI or whenever a situation involves a question they just instantly say "but the AI says!" It's actually so annoying.
I remember early in AI development, one metric people looked at was how often the AI played the same move as a pro. I think that number ends up around 30%, which means that around 70% of the moves will be something different which will probably be at least -0,5 % :-) according to the AI
@@mathisisaksen2215 I disagree with the idea that anything but the top move loses something. Especially in the opening there are many moves that are equally good. Also -0.5% wr /
@@katrain I said that of the 70% of moves that do not coincide with the AI most, not all, will be considered a small loss - and I agree that such a small a difference can be considered statistical noise :). One can just let their computer run a little while to see how the evaluation might change to observe the uncertainty in the numbers :)
As a lowly 5 kyu, I find AI a bit tricky to use.. I run into this problem where my opponent and I both look at a group and think "that's probably fine." But it's not fine, and the AI spends the next 100 or so moves swinging back and forth because whom-ever's turn it is is the obvious winner :D
Thanks for posting this, it's really great to have some advice for how to maximize the utility of the analysis tools that are available to us. Something as simple as having questions that you want an answer to makes a big difference in understanding and learning, vs. just looking at inflection points in the win rate graph. Also really appreciate the emphasis on ignoring the < 5 point mistakes when reviewing
This sounds solid. It is an extension of the principal that it is important to have a plan, because even if that plan fails at least you know what failed and how.
There is a big point that is too obvious to say. Do not try to emulate the AI. You will never be able to do that. There, said it.
Nonsense. Plenty of AI innovations initially dismissed as incomprehensible are being studied by human players and incorporated intelligibly into human games even at the kyu level. I can give plenty of examples on request.
A more accurate phrasing of your advice would be, "try to understand what you emulate (but also be aware that some lessons only come with practice)." If you get a piece of advice from an AI that resists all attempts to be understood, consider it provisional if you don't forget it altogether; if you get a piece of advice from an old Japanese fuseki book that resists all attempts to be understood (and as someone who's read plenty of Go books before AI became superhuman, trust me there is no shortage of under-explained, inscrutable, even inaccurate pro advice), consider it provisional if you don't forget it altogether. Blindly emulating humans is no better than blindly emulating AI (and if you do, at least the latter is objectively stronger, less prone to blind spots and overconfidence, and infinitely patient with you).
@@Enuchful That's a fair response.
This was actually pretty interesting :)
Great points. Thank you for the video.
I created an account using your link to support you as usual especially as it's a paid promotion, but I was looking forward for basics :'( Guess I'll have to patiently wait for next week 😁
Thumbs up for this video by the way :-) thanks for your input on AI reviews.
I'm still waiting for a followup to your asmr video.
Also I got to see the first game live, so it's interesting to see the video you talked about.
Why wouldn't you mention KaTrain? It's free, and uses the very strong KataGo.
I wanted to stick with something anyone could access. As stated in the video, some people cant run katrain on their device. Also I liked the features of ai sensei, which also has a free option.
One typically doesn't mention the competition in a sponsored video, and as a free tool, I can't really sponsor much.
Regardless, AI sensei is a great tool if you prefer something web-based, and if dwyrin found any features missing in particular in KaTrain I'd be happy to hear it.
AIsensei finally getting the love it deserves
I'm a site contributor on OGS, and I get breddy gud AI reviews there. So do my opponents, regardless of if they're donors.
Is there an editing mistake at the end?
a content creator never makes mistakes. they create precisely what they mean to!
Very good video. Thanka
How can I self-review WITHOUT the aid of an AI? Especially if I lost the game...
By going back and taking the time to read out the variations you couldn't read out in game, as well as keeping basic principles in mind (make good shape, attack weak groups, play big moves) to see which of your moves were slow.
Well, usually during your games you'll have OH SHIT moments, where your opponent separates two of your groups, or kills a group you thought was alive, or something other that's huge for him. Going back to those and seeing what you could have done to avoid it is a good start.
Other things to keep in mind:
My opponent did X and I responded locally. Was X really sente for him?
I got sente and did X. Should I have done something else?
These are great replies, thanks to both of you.
Great video, but Dwyrin is so rude! He cut Dwyrin off mid-sentence at the end of the video!
too bad I registered with them yesterday... would have loved to give you that small support by using your link. :(
Confirmed, BattsGo is BotsGo
01101000 01101111 01110111 00100000 01110010 01110101 01100100 01100101
01101110 01100001 01101110 01101001 00111111
I think people misunderstand how to use AI way too often they borderline misunderstand it so bad that they try to become the AI or whenever a situation involves a question they just instantly say "but the AI says!" It's actually so annoying.
I wonder how many -0.5% mistakes a 9p would typically make in a real time game.
I remember early in AI development, one metric people looked at was how often the AI played the same move as a pro. I think that number ends up around 30%, which means that around 70% of the moves will be something different which will probably be at least -0,5 % :-) according to the AI
@@mathisisaksen2215 I disagree with the idea that anything but the top move loses something. Especially in the opening there are many moves that are equally good. Also -0.5% wr /
@@katrain I said that of the 70% of moves that do not coincide with the AI most, not all, will be considered a small loss - and I agree that such a small a difference can be considered statistical noise :). One can just let their computer run a little while to see how the evaluation might change to observe the uncertainty in the numbers :)
🐙