That’s not remotely true, is it?
If Monte Carlo Tree Search is not reading, what is it?
What Leela Zero can’t do is extrapolate. If the search stops before the ladder bottoms out, she doesn’t go “Oh, I see that this will end over there”.
That’s very different to “no reading”.
Happy to be corrected, just checking my whole basic understanding of the thing! 
So I agree that MCTS is “reading”, but when I stated LZ is “high dan” without any reading, I’m specifically refering to a recent test some people (including myself) did where we test LZ on one single playout (essentially no MCTS and thus no ‘reading’) and it reached about 4D or 5D… so perhaps not “high dan” but certainly “mid dan”… and its network is still improvement and I’m sure it can reach higher without reading once it gets to 20 blocks. BTW, in AGZ paper, the AGZ raw network without any ‘reading’ is already itself a pro level.
Now of course this may not translate all that perfectly to the human play, but to some extent if we can compare the “network” to human intuition and “MCTS/po” to “reading”, then it would seem that the analogy is a pro that didn’t read would still play stronger than a low dan player that did read. To complete the analogy, “reading” is like upping the hardware while “intuition” is like upping the network. As a beginner one can certainly improve via any numbber of ways since one hasn’t reached deminishing returns and everything is low hanging fruit, so therefore I would like to think my preference is to instill and improvement my human intuition network first, rather than improve my reading/brute force skills. I find the former to be more fun than the latter.
I see - thanks for elaborating, I had no idea the network was that good.
However, it has the intuition from seeing millions of games right? Maybe when a human has reviewed that many, that person too would have that good intuition?
Yeah I see what you are saying. Its like the millions of self playing games somehow condensed and encoded itself as ‘intuition’ when if one thinks about it, a five year old child can be taught the game in one afternoon and within a couple of games could play fairly good… Yeah take Leela Zero for example, already it is at 7 million games. That’s probably more than all of the human games in the history of Go combined… and certainly orders of magnitude more than any one pro could ever cram into an entire lifetime… so from that perspective the Deep learning really isn’t as much “intuition” as it is still another form of “brute forcing”, just here instead of MCTS brute forcing, we are brute forcing to create the “intuition” network instead. A true superintelligence would be able to pick up on the game using LESS than the total number of games that is required for someone to go from zero to top pro. And in that metric, we are no where near that sort of AGI. (general artificial intelligence)
For me, as a beginner, I think training that sort of “inutition” is better than raw reading skills or just tactics.
Reaching dan level without reading practice is a fantasy.
So is reaching SDK.
I don’t think pushing tsumego on beginners is a good idea. It’s pros who need to read hundred moves ahead so they do tsumego every waking moments of their lives. We aren’t pros. For most kyu games only short sequences in important places require reading. And often times they are simpler than convoluted tsumego (e.g. can this one space jump be cut?). Plus beginners often don’t even know what to read because for them each move is just as good as any other.
Without reading you won’t get anywhere but with a little reading you can get much further than it seems. In the end intuition and experience are more important. SDKs don’t need to read L-group, they know it’s dead. They don’t need to read 3-3 invasion because they’ve done it many times already. They know to bamboo joint instead of connecting in a string.
To train intuition it’s probably best to play and look at commentated strong player’s games. I’d still recommend doing simple tsumego for recreational purposes but not as silver bullet for all your problems.
I agree.
This point is made by the Back To Basics series. There is a lot of progress you can make before you need to be able to read out detailed fights.
It’s one of the dangers of “general advice”. Appropriate advice for Dan, SDK, DDK and TPK is likely to be significantly different. (I also think that at the distance of Dan from DDK and TPK it can possibly be hard to even make out the difference 
)
My own experience has been that a little tsumego started benefiting me at about 17k. FWIW.
There are 1 move or so tsumegos…Doing those is not a tall order. L- Group might be dead, but do you know how to die with the most liberties in a capturing race? Are you sure that L group doesn’t have that extra hane connect move on the first that is Sente that now makes it alive? Are you sure that 3-3 group is truly alive/dead? Many people only learn those 1 or 2 variations and they think they’ve mastered 3-3. Making light of these things because you think they are the same is insulting to the game and is truly a way for SDKs to stay SDK.
That makes perfect sense to me: I have come to appreciate that to move up into SDK I will need to do more tsumego.
On the other hand, this thread was originally about 23ks
But you can get to SDK without all these complications, can’t you? And still, things like how to die with most liberties sounds like a question of experience rather than reading. Of course, tsumego is a way to gain experience in local skirmishes but not the only way.
1 move tsumego are fine indeed.
Poor game of go. It’s never going to be the same again 
What I got out of the Back to Basics series was in fact the exact opposite - that reading is very important, and intuition without reading is bad! (And the converse too.) The hint lies in the fact that when dwyrin says something along the lines of “now we’re safe, time to make a big move” or “the cut doesn’t work, so we have sente”, he’s actually reading out a little bit to judge when it’s alright to step away. When there’s nothing urgent and concrete, then we can rely on intuition. For examples where intuition alone is used… the “This is Fine” video of the series shows what can go wrong.
I say all this even though I’m terrible at go because it’s exactly the same in chess (where I’m much less bad). No point discussing positional play if free pieces are given away. Likewise in go, no point taking big points if all the stones just get cut to shreds because of not reading out cuts and invasions. Improvement is had by improving both aspects at the same time - intution and tactics.
I agree that you have to read to assess this. I guess then I am forced to change my earlier statement to “deep reading is not required”. It seemed obvious that of course you have to read at least your opponent’s next move and your response to that. So my earlier statement was an “overstatement” or “over-generalisation”.
HOWEVER
This is not an example of where reading is required. In these instances, my appraisal is that Dwyrin is looking at the shape and saying “it looks safe” because of the shape itself. Unlike the cut, he doesn’t read “what if this and what if that”. Rather, good shape provides a shortcut against exactly having to do this. If you have good shape, you don’t have to read every possible assault on it.
Generally speaking, Dwyrin would frequently assert that he is not reading when he plays basic, and even from time to time apologised when he caught himself doing it. He frequently says “I could do this, but I’m not going to because it requires reading to know that, so I am doing this basic thing instead”.
GaJ
I’m on iOS and I’m not sure of Tsumego Pro for iOS is by the same guy as Tsumego Pro for Android so my comments about that app you are currently using probably don’t apply. That said, GoProblems.com has an android and web version as well as iOS.
https://play.google.com/store/apps/details?id=com.adum.gpandroid&hl=en
I’m not sure how you really expect to train your intuitions without any reading? I’m certainly not all that strong, but realistically whatever intuitions I have (suspect as they might be at supposed 5k) are the result of tons of reading practice over time, and at the moment I tend to believe that the most painful blanks in my intuitions are basically where I haven’t done enough life and death problems / reading.
(Also, I’d caution you to be pretty suspicious of any “intuition” trained at a ddk skill level, you might well have to unlearn it all eventually – I know that my sdk intuitions definitely get me into trouble sometimes.)
I think this is a very good point. Especially true if playing against weaker opponents and winning too many games as a beginner.
Not really no. I don’t believe there is truly intuitive style of play outside of 1 second blitz where you truly have no time to read out. Otherwise I believe intuition still involves 1-2 moves of reading ahead. Experience for liberty count? No…you manually read it out.
Anyway kind of off track now, but just saying that I highly disagree with your ‘intuition’ based play being able to get to SDK. There is certainly still some reading involved.The only true intuitive play are centre-based plays and the fuseki stage of the game where it is humanly impossible to read out all the plays.
Weak people who get to SDK do so not bothering to do joseki but still relatively ok at reading to take advantage of their overplays. Unless you’re a robot who has auto-calculated the entire game in your subconscious can you truly play at that high a level with just intuition. Or of course just being lucky that your opponent messed up more than you in blitz when you both don’t read.
Pretty much the harsh truth I needed to hear. As a millennial (I know, I know — we’re terrible), I just want to be good at Go already without actually putting in the work.
It doesn’t feel like work if you enjoy doing it.
Work is work 
But ultimately for most of us Go is about having fun, indeed.
