That was a good development on how rumors might become self-sustaining, when not checking on sources of the bits of information that they might have started from, without context, one has a tendency to rely on others words (social animals that we are), and a common clamor, tends to have its own power of persuasion.
While I agree that borrowing engine glasses to view the whole of chess might induce such tunnel vision, or tunnel style or other alternate turn short term sequences blinders against strategic or long term visible static effects of early position choices (the ones after the moves), I think that for SF in particular, since there is no long term concept but a deepening beyond our abilities to do the same of short term calculation, which when short as humans we tend to call tactical, we would be able to extract more human valid information from the deep leafs actual iota of positional hand-crafted programming that SF might have.
It would not matter if that particular path was not "stable", the current score for the position we have in our face, is from that last position of that first PV. Then at least we might start as a population that tend to take SF like a god of chess, to look at the SF goggles with our own goggles..
And even if SF hand-crafted has been dominated by old fishtest optimization to be mostly listening to material imbalances at the leaf positions, with some nested to that some modulation for positional theory concepts thrown in the soup, each with own parameters fishtest optimized one at a time given frozen previously optimized concepts parameters, including the very first material count concept). These leafs, at depth beyond our calculation abilities, are the positions being actually looked at by SF, not really the ones that 16 "safe" plies that lichess truncate SF PVs with.
I think that at least the example where one ply made the big swing SF internal search novelty discovery (in its chess truncated trees model of full chess). Could be humanly chessly understood if it were systematic that lichess would provide for the position were the actual position evaluation (from complete position information) that was made and which finally gave the score at current position..
Long PV "stability" of path be damned.. The human visibility of the leaf features in parallel with the score, itsinformative for human goggling of SF googling (it search tree). We might get to educate ourselves about why SF misses something that we could otherwise already see or learn to see at current position, but understanding that SF is scanning way ahead for things it was programmed to see.. But what are those, let's find out. let this be part of SF feedback for better less gullible use of it..
While I agree that borrowing engine glasses to view the whole of chess might induce such tunnel vision, or tunnel style or other alternate turn short term sequences blinders against strategic or long term visible static effects of early position choices (the ones after the moves), I think that for SF in particular, since there is no long term concept but a deepening beyond our abilities to do the same of short term calculation, which when short as humans we tend to call tactical, we would be able to extract more human valid information from the deep leafs actual iota of positional hand-crafted programming that SF might have.
It would not matter if that particular path was not "stable", the current score for the position we have in our face, is from that last position of that first PV. Then at least we might start as a population that tend to take SF like a god of chess, to look at the SF goggles with our own goggles..
And even if SF hand-crafted has been dominated by old fishtest optimization to be mostly listening to material imbalances at the leaf positions, with some nested to that some modulation for positional theory concepts thrown in the soup, each with own parameters fishtest optimized one at a time given frozen previously optimized concepts parameters, including the very first material count concept). These leafs, at depth beyond our calculation abilities, are the positions being actually looked at by SF, not really the ones that 16 "safe" plies that lichess truncate SF PVs with.
I think that at least the example where one ply made the big swing SF internal search novelty discovery (in its chess truncated trees model of full chess). Could be humanly chessly understood if it were systematic that lichess would provide for the position were the actual position evaluation (from complete position information) that was made and which finally gave the score at current position..
Long PV "stability" of path be damned.. The human visibility of the leaf features in parallel with the score, itsinformative for human goggling of SF googling (it search tree). We might get to educate ourselves about why SF misses something that we could otherwise already see or learn to see at current position, but understanding that SF is scanning way ahead for things it was programmed to see.. But what are those, let's find out. let this be part of SF feedback for better less gullible use of it..
