Where are the funs of Leela?

[Nay Lin Tun]

End of an era was already spread like viral!
OP think that it is me who is exaggerating. It is partially true in this forum. But the news already spread like wildfire in media sources like " youtube, reddit, chessnews sites", and also in "tcec and chess.com twitch chat"

One of reasons is that people want to hear interesting things and media spread/exaggerate this.

Honestly, I personally still think Leela is +20 to +30 elo above SF in TCEC conditions.


https://chess24.com/en/read/news/a-new ... stockfish

leelachesszero-stockfish-net.jpg

[corres]

End of an era was already spread like viral!
OP think that it is me who is exaggerating. It is partially true in this forum. But the news already spread like wildfire in media sources like " youtube, reddit, chessnews sites", and also in "tcec and chess.com twitch chat"
One of reasons is that people want to hear interesting things and media spread/exaggerate this.

From peoples who just the user of Stockfish and Leela this is a natural reactions.
But from peoples who are "near to the blaze" and think (with more or less establishing) he is an expert would manifest about the theme with more low-key manner - even if he is believer of Leela.
At least this is wished by the correctness.

Honestly, I personally still think Leela is +20 to +30 elo above SF in TCEC conditions.

Naturally everybody have right to their opinion.
But the value of every opinions are determined by how the opinions reflect the facts.
The "TCEC conditions" belongs to not only the used hardware but the system of the competition, the participants of the TCEC, the used opening book and the used Time Control also. Maybe you are right in a paired competition Leela is stronger with some Elo on the hardware of TCEC. But as the Champion Bonus proves in an another "TCEC conditions" Leela produces rather weak achievement.
Moreover "End of Era" is very sonorous statement but it is a non-established opinion as the above posts prove.

[carldaman]

As someone else suggested, the 'Dawn of a New Era' would be a better and somewhat less sensational way to express the same thought.

[Nay Lin Tun]

End of an era was already spread like viral!
OP think that it is me who is exaggerating. It is partially true in this forum. But the news already spread like wildfire in media sources like " youtube, reddit, chessnews sites", and also in "tcec and chess.com twitch chat"
One of reasons is that people want to hear interesting things and media spread/exaggerate this.

From peoples who just the user of Stockfish and Leela this is a natural reactions.
But from peoples who are "near to the blaze" and think (with more or less establishing) he is an expert would manifest about the theme with more low-key manner - even if he is believer of Leela.
At least this is wished by the correctness.

Honestly, I personally still think Leela is +20 to +30 elo above SF in TCEC conditions.

Naturally everybody have right to their opinion.
But the value of every opinions are determined by how the opinions reflect the facts.
The "TCEC conditions" belongs to not only the used hardware but the system of the competition, the participants of the TCEC, the used opening book and the used Time Control also. Maybe you are right in a paired competition Leela is stronger with some Elo on the hardware of TCEC. But as the Champion Bonus proves in an another "TCEC conditions" Leela produces rather weak achievement.
Moreover "End of Era" is very sonorous statement but it is a non-established opinion as the above posts prove.

Most of my opinions are based on results of leela tests on discord. As Tcec sample games are very small, the tournment results can varies from one to another.

In Leela discord , before Sufi 14 and Sufi 15, there were extensive amounts of games tested with exact SF vs Leela ratio in Tcec, around 5000 to 10, 000 games before s14 and 20,000+ games before S15.


Afaik, from one tester' report from approx 8000 games from Tcec opening book , the results says S15 Leela was +17 elo above SF, and S14 Leela was -20 elo below SF.

In real tcec final, S14 Leela overperformed and ended with -5 elo below SF and S15 Leela also overperformed and ended with +25 elo.

I think there are more than 500,000 games tested on Leela vs SF from various testers in 2019 alone.( some are custom testing conditions and some are exact simulation of Tcec condition)

With those data, I am still confident that Leela will again beat SF in S16 with similar results from S15.( note- there is approx +10 elo gain from SMP SF but newer Leela nets also gain +5 to +10 elo as of today)

[jp]

In Leela discord , before Sufi 14 and Sufi 15, there were extensive amounts of games tested with exact SF vs Leela ratio in Tcec, around 5000 to 10, 000 games before s14 and 20,000+ games before S15.

Those tests must have used much smaller nodes per move, though, if there were so many games. That would then be just bullet game testing. It's not just the ratio that matters.

[MikeGL]

I agree these NN are very strong. But I did a quick visit at CCC site for the currrent tourn, and was suprised to see Brainfish was demolishing Lc0 left and right. Don't know what's wrong but looks like it is vengeance time for A/B side.

[mclane]

To me it is completely unimportant how lc0 plays Vs. Stockfish.
I put lc0 on my SLOWEST android device and look how it plays chess games against OTHER engines.

Lc0 Vs. Stockfish, both running impressive hardware is IMO boring.
I am more interested in seeing lc0 play strategic against AB programs who play mostly tactical.

I find it extremely funny to see how an engine generates ELO with strategy and absence of tactical skills.

On my SLOWEST android device lc0 makes 10-15 NPS.

[mclane]

[pgn] [Event "40/120"] [Site "SCW"] [Date "2019.06.03"] [Round "1"] [White "LC0 Android Element"] [Black "MCE The King 2.61"] [ECO "C41"] [Result "1/2-1/2"] 1. e4 e5 2. Nf3 d6 3. d4 Nd7 4. Bc4 c6 5. Ng5 Nh6 6. O-O b5 7. Be2 Be7 8. Nf3 Ng4 9. a4 b4 10. c3 a5 11. cxb4 axb4 12. Qc2 Ba6 13. Bxa6 Rxa6 14. h3 Ngf6 15. dxe5 Nxe5 16. Nxe5 dxe5 17. Be3 O-O 18. Nd2 c5 19. Nc4 Qa8 20. f3 Re6 21. a5 Rb8 22. Rfd1 Qa7 23. b3 Nd7 24. Rd5 Qc7 25. Bf2 Nf8 26. Qd3 Rb7 27. Ne3 Ng6 28. Nf5 Bf8 29. g3 Ra7 30. Rd8 Rea6 31. Qd5 h6 32. h4 h5 33. Kg2 Rxa5 34. Rxa5 Qxa5 35. Bxc5 Qa2+ 36. Bf2 Qb2 37. Qd2 Qxd2 38. Rxd2 Rc7 39. Rd3 Rc1 40. Ne3 f6 41. Nd5 Rc2 42. Kf1 Rb2 43. Be3 Rb1+ 44. Ke2 Rb2+ 45. Kf1 Rb1+ 46. Kf2 Rb2+ 47. Ke1 Rb1+ 48. Kd2 Rb2+ 49. Ke1 Rb1+ 50. Ke2 Rb2+ 51. Bd2 Bc5 52. Kf1 Rb1+ 53. Ke2 Rb2 54. Kf1 Rb1+ 55. Ke2 Rb2 1/2-1/2 [Event "40/120"] [Site "SCW"] [Date "2019.06.03"] [Round "1"] [White "MCE The King 2.61"] [Black "LC0 Android Element"] [ECO "B02"] [Result "0-1"] 1. e4 Nf6 2. Nc3 d5 3. exd5 Nxd5 4. Bc4 Nb6 5. Bb3 c5 6. Qh5 c4 7. Bxc4 Nxc4 8. Qb5+ Nc6 9. Qxc4 e5 10. Nf3 Be7 11. O-O O-O 12. Qe4 f5 13. Qc4+ Kh8 14. d3 Bf6 15. Re1 h6 16. Bd2 Re8 17. Qb3 a5 18. Nd5 Be6 19. c4 Bxd5 20. cxd5 Nb4 21. Bxb4 axb4 22. d4 e4 23. Ne5 Bxe5 24. dxe5 Rxe5 25. Rad1 Qd6 26. Rd4 Rd8 27. Red1 f4 28. Re1 Qg6 29. Rxb4 Rdxd5 30. Rb6 Qh5 31. Rxb7 f3 32. Qc4 Kh7 33. Rb3 Rc5 34. Qf1 Qg6 35. g3 Qe6 36. Rc1 Rcd5 37. Rcc3 Rd2 38. Re3 Qd5 39. h3 h5 40. h4 Rd4 41. Kh2 Rd1 42. Re1 Rd2 43. Kg1 Qd4 44. Ree3 Qd5 45. Kh1 Rd4 46. Re1 Qe6 47. Kg1 g5 48. Rb7+ Kg6 49. Rd1 Rdd5 50. Rxd5 Rxd5 51. b4 e3 52. Qb1+ Rf5 53. Qb2 Rf6 54. Qc2+ Rf5 55. Qb2 Rf6 56. hxg5 e2 57. Qb1+ Rf5 58. Qe1 h4 59. a4 hxg3 60. Rb8 0-1 [Event "40/120"] [Site "SCW"] [Date "2019.06.15"] [Round "1"] [White "LC0 Android Element"] [Black "MCE The King 2.61"] [Result "1/2-1/2"] 1. f4 g6 2. e4 d5 3. exd5 Nf6 4. d4 Qxd5 5. Nf3 Bg7 6. Nc3 Qd6 7. Bc4 O-O 8. O-O a6 9. Ne5 Nc6 10. Be3 e6 11. a4 b6 12. Be2 Bb7 13. Bf3 Nd5 14. Nxd5 exd5 15. Bf2 Rfe8 16. c3 Rad8 17. Re1 f6 18. Nd3 Na5 19. f5 gxf5 20. Bg3 Rxe1+ 21. Qxe1 Qd7 22. Bf4 Nc4 23. Qg3 Nd6 24. Bxd6 Qxd6 25. Nf4 c6 26. Re1 Bc8 27. Qh4 Bd7 28. g3 Re8 29. Rxe8+ Bxe8 30. Qh3 Bd7 31. Be2 Qe7 32. Kf2 Qe4 33. Bd3 Qh1 34. Bxf5 Bxf5 35. Qxf5 Qxh2+ 36. Ng2 Qh6 37. Qe6+ Kf8 38. Nf4 Qh2+ 39. Kf3 Qh1+ 40. Kf2 Qh2+ 41. Ng2 Qh6 42. Qxc6 Qd2+ 43. Kg1 Qd1+ 44. Kh2 Qh5+ 45. Nh4 Qe2+ 46. Kg1 Qd1+ 47. Kf2 Qc2+ 48. Kf3 Qd3+ 49. Kf2 Qd2+ 50. Kf3 Qd3+ 51. Kf2 Qc2+ 52. Kf1 Qd1+ 53. Kg2 Qe2+ 54. Kg1 Qe1+ 55. Kg2 Qe2+ 56. Kh3 Qf1+ 57. Kg4 Qe2+ 58. Kh3 Qf1+ 1/2-1/2 [/pgn]

[M ANSARI]

Can someone fix or edit the "Where are the funs of Leela". I think the person meant to write "fans" rather than "funs". Is it possible to edit your titles after posting?

[Laskos]

I agree these NN are very strong. But I did a quick visit at CCC site for the currrent tourn, and was suprised to see Brainfish was demolishing Lc0 left and right. Don't know what's wrong but looks like it is vengeance time for A/B side.

I had a look now too. You have to be aware of the conditions. Right now they are playing starting from the move 22 (!) of some important game of the past. So, right from the middlegame, completely missing the opening and early midgame. The games are rapid, not LTC. BF is on 90 real cores, Lc0 is on 4x 2080ti GPUs (and doesn't scale well with them).

All in all, in these conditions favoring BF, especially that the openings are completely skipped, +7 -3 result for BF is expected.