I vaguely recall either Michael Redmond or TelegraphGo saying on one of their youtube videos that top AI programs have shown us that checking extensions aren’t as valuable as we previously believed, unless they have an extremely strong follow-up if ignored (or some other narrow caveat or rare circumstance that must be met). However, my memory is about as unreliably decrepit as a 1990’s vhs tape that has been recorded over one too many times.
Does anyone remember hearing either of these 2 go youtubers mention something about this or remember their opinion on checking extensions after the ai go revolution? Or does anyone happen to know what strong ai engines like Katago think about checking extensions overall? This is the (slightly unrelated) youtube video which sparked my question about checking extensions, although this example is likely better defined as a counter-pincer.
Yes, like many side-moves we overvalued them before the AI-revolution. Using my Leela Zero Opening Gospel there are no side moves in it, so you should’t play a checking extension if any of the corner moves in it are still available, e.g. don’t play either of them on the right side here, but do something to the left corners (even more so if they were 3-4s not 4-4s):
Uberdude, is the right interpretation of your “opening gospel” that you should only play side moves in the opening when they are the standard final move of common corner joseki? Or is it instead that you should omit the final move(s) in all the previously-common corner joseki if they end in an extension to the side?
Maybe not your prototypical checking extension of being one space from their side move threatening an invasion / pincer follow-up, but it’s a side extension preventing opponent’s desired (but not until later) extension in front of shimari, with follow up of a further extension with further follow ups inside the shimari. But it was too early, side before corner, so Master punished it by cutting in the corner which was bigger and where white should have played.
The first part of the gospel is to play urgent moves before big moves (nothing revolutionary there!), and gives examples of common urgent moves in the early opening with their priority order, and then goes onto the more fuseki-style plonking stones in non-urgent big areas priority order. Joseki are usually an urgent local fight, so it’s urgent to keep on playing the joseki and the gospel doesn’t tell you not to do that. Obviously there’s an element of judgement to “is this urgent and I should keep playing locally, or go back to gospel priority-order stone plonking?”, it can’t make you play an AI-level opening with zero thought or go skill! But yes, it does emphasise the importance of getting sente to get back to those big corner moves so if the extension isn’t needed to stabilise a group maybe you can more strongly consider it. Or here’s an example where before making the simple extension to A/B (still excellent moves), you could do the AI-popularised attach into the big shimari as a bit of a probe: if you can get some stones here in sente maybe you can be more efficient that the plain extensions, and it also inteferes with black doing the checking extension of C.
P.S. I just checked the above with AI and actually it says q16 is an overplay because black should resist with q15 outside hane (not obedient q17 inside hane), and better to just do simple extension
P.S. Although I called it the Leela Zero Opening Gospel (because that was the top AI available at the time I studied a lot with to formulate it), those 60 Master games and Michael Redmond’s excellent commentary series were also an input, and indeed in many of the games the pros fell behind by move 30, but wouldn’t have had they followed the gospel.
tygemfox/10-2016.12.30-Magist-Meng_Tailing.sgf for example illustrates that Master agrees that “enclose your 3-4 corner” is bigger than “split a side” / “approach top right”, contrary to prevailing human theory fashion at the time. Redmond talks about this pattern of Master making a patient move 6 in his videos.
