AI Sensei

You play a game of Go. You review it with AI. You see where you went wrong. You nod thoughtfully, close the review, and then... make the exact same mistake two games later.

Sound familiar? You're not alone. It's probably the most common frustration in Go study.

The Gap Between Knowing and Doing

Understanding a mistake and actually fixing it are very different things. When you review a game and see that move 63 should have been a shoulder hit instead of an attachment, that's useful information. But information alone doesn't build habits.

Think about how other skills work. A tennis player doesn't just watch video of their bad serves — they go practice serves. A musician doesn't just listen to the passage they stumbled on — they drill it until it's smooth.

Go players, traditionally, haven't had a clean way to do this. You review the game, understand the mistake, and then hope it sticks when a similar position comes up.

What Training Mode Actually Does

AI Sensei's Training Mode closes that gap. Here's how it works:

1. You upload and review a game — AI Sensei analyzes your game and identifies your key mistakes
2. You mark positions worth practicing — hit "Remember" on any position during your review
3. The positions become practice problems — they get saved to your personal training set
4. You practice with smart scheduling — problems you get wrong come back more often, ones you nail gradually fade out. It's like flashcards, but for Go positions.

The problems aren't from a generic problem set. They're from your actual games. The positions you struggled with. The mistakes you really made. That's what makes it different.

Why Your Own Mistakes Are the Best Practice Material

Generic tsumego collections are great for building reading ability. But they have a limitation: they don't know what you specifically get wrong.

Maybe you always misread the timing of a 3-3 invasion. Maybe you consistently choose the wrong direction for a splitting attack. Maybe your endgame technique is solid but you collapse under pressure in capturing races.

Training Mode targets exactly those weak spots. Because the problems come from positions where you actually went wrong, you're training the specific patterns your game needs most.

It's like having a coach who says "I noticed you keep making this type of mistake — let's drill it until you get it right." Except the coach has analyzed every single one of your games and never forgets.

The Practice Loop

The most effective way to use Training Mode is as part of a regular review cycle:

Play → Review → Remember → Train → Play again

1. Play your game (on any server — Pandanet, OGS, Tygem, WBaduk, wherever)
2. Upload to AI Sensei and review your key mistakes
3. Hit "Remember" on the positions that matter most to you
4. Practice your saved positions in Training Mode
5. Play your next game with the corrections fresh in your mind

Over time, the mistakes you keep practicing are the ones that stop showing up in your games. That's not a theory — it's how pattern recognition works. Repeat the correct response enough times, and it becomes your instinct.

Not Just for Strong Players

A common misconception is that AI review (and by extension, Training Mode) is only useful for dan-level players. In reality, it works at every level — the mistakes are just different.

A 15 kyu might save capturing races and basic life-and-death positions they missed in their games. A 5 kyu might practice direction of play and when to tenuki. A 3 dan might drill subtle endgame sequences and ko timing.

The AI doesn't care what your level is. It finds where you lost points, and you choose which positions to train. The practice is always matched to your actual level because it comes from your actual games.

No Other Tool Does This

Most Go analysis tools stop at showing you the review. They'll tell you what went wrong and what the AI would have played. That's valuable, and plenty of tools do it well.

But none of them take the next step of turning those specific mistakes into practice problems you can drill on a smart schedule. It's the difference between diagnosis and treatment.

AI Sensei's Training Mode is, as far as we know, the only tool in the Go world that completes this loop. Review your game, save the key moments, then practice them until they stick.

[IMAGE: Simple visual showing the Play → Review → Remember → Train → Play cycle, with a Go board in the center]

Try It

Upload a game at ai-sensei.com and review it. When you spot a mistake that feels important — one you know you'd make again — hit "Remember." After reviewing a few games, open Training Mode and start practicing.

You might be surprised how much harder it is to find the correct move under pressure than it was to understand it in review.

That gap between "I understand" and "I can do it" — that's where improvement lives. And Training Mode is how you close it.

Training Mode is available on all AI Sensei subscription plans. Upload your first game for free at ai-sensei.com.

Not long ago, studying Go meant replaying your game from memory, hoping your teacher would spot the key moments, and working through problems that may or may not match the mistakes you actually make in games.

AI changed all of that.

The Before: How Go Players Used to Study

Traditional Go study had a well-worn path: play games, review them (ideally with a stronger player), study joseki and tesuji from books, and solve tsumego problems. It works — generations of strong players prove that. But it has real limitations.

Finding a teacher who can review your games regularly is hard. Even when you do, they have limited time. They might focus on the opening when your real problem was a middle game fight. And the feedback cycle is slow — you play today, get reviewed next week, and by then the details are hazy.

For most amateur players, self-review meant staring at the board and thinking "I know something went wrong around move 47, but I'm not sure what." Sound familiar?

The AI Revolution: From Guessing to Knowing

When AI engines like KataGo arrived, they brought something genuinely new: the ability to evaluate every single move in a game with superhuman accuracy.

Suddenly, you don't have to guess where things went wrong. The AI can tell you:

• Which moves lost the most points — so you focus on what actually mattered
• What you should have played instead — with concrete variations, not vague advice
• How the game's win rate shifted — so you can see the flow of the game at a glance

This isn't just faster than human review. It's a fundamentally different way to study. Instead of working from general principles ("I think my direction of play was wrong"), you're working from precise data ("Move 47 lost 8 points; here's the sequence that holds the lead").

But Raw AI Output Isn't Enough

Here's the thing about KataGo and similar engines: they're incredibly strong, but they weren't designed to teach. Running KataGo locally means dealing with setup, GPU requirements, and an interface that shows you everything without filtering what matters.

For a 5 kyu player reviewing a game, seeing AI suggestions on all 250 moves isn't helpful — it's overwhelming. The real question isn't "what does the AI think about every move?" It's "which of my moves actually mattered, and what should I learn from them?"

That's the gap between raw AI analysis and actual learning.

Focused Study: Finding What Matters Most

The most effective way to use AI for Go study is to focus on your biggest mistakes — the moves where you lost the most ground. Not every inaccuracy, not every suboptimal play, but the moments where the game actually turned.

This is exactly what AI Sensei is built to do. When you upload a game, it highlights your key mistakes and shows you what the AI would have played instead. You can quickly scan through a game's critical moments without getting lost in noise.

Think of it as a filter between the raw AI and your brain. The engine does the heavy computation. The interface shows you what's worth studying.

From Review to Practice: Closing the Loop

Reviewing a game and understanding your mistakes is valuable. But there's a step most players skip: actually practicing the correct moves.

It's like reading a book about swimming versus getting in the pool. You can understand intellectually what you should have played, but unless you practice the right response, you'll make the same mistake next time.

That's why AI Sensei's Training Mode exists. As you review your game, you can mark any position worth practicing. These get saved as problems that come back automatically — the ones you get wrong show up more often, and the ones you nail gradually fade out. It's like flashcards, but for Go positions.

No other Go analysis tool does this. It's the difference between knowing what went wrong and actually training yourself to get it right.

No GPU Required: AI Analysis for Everyone

One barrier to AI-powered study has always been hardware. Running KataGo locally requires a decent GPU and some technical setup. That's fine for tech-savvy dan players, but it locks out most of the Go community.

Cloud-based analysis removes that barrier entirely. Upload your game file — SGF from OGS, GIB from Tygem, NGF from WBaduk, UGF from Pandanet — and the analysis runs on powerful remote servers. You get results in minutes, viewable from any device with a browser.

No installation. No GPU. Just your game and a few minutes of patience.

What AI Can't Replace

It's worth being honest about what AI review doesn't do.

AI won't explain why a move is good in human terms. It won't say "this move is important because it creates influence toward the center while limiting your opponent's base." It gives you the right move and the reading to prove it, but the conceptual understanding still needs to come from human sources — teachers, books, stronger friends, your own reflection.

AI review is also not a substitute for playing. The best study routine combines playing games, reviewing them with AI, practicing key positions, and building conceptual understanding through traditional study. AI just makes the review step dramatically more efficient.

Getting Started

If you haven't tried AI game review yet, the easiest way to start is to upload a recent game to AI Sensei. It's free for basic analysis — the AI's positional judgement is superhuman, though it may occasionally misread complex tactical positions. Even so, it'll spot your biggest mistakes reliably.

Just grab an SGF file (or GIB, NGF, UGF — most server formats work), upload it, and look at the moves highlighted with the biggest point losses. Start there. That's where the learning is.

The AI won't make you stronger by itself. But it will show you exactly where to focus your effort — and that's how improvement happens.

AI Sensei uses KataGo to analyze Go games in the cloud. Free analysis is available at ai-sensei.com.

After the last update, we got a lot of feedback from users who prefer the old input method on touch screens. So we added options to customize the behavior more and changed the defaults back to how it was before.

On touch, there's also now a confirm dialog for important moves like when playing against an AI. There's a setting for this, too.