Since yesterday my bots are crashing with the error
e[33mApr 17 18:26:02 main.ts:143 Disconnected from servere[39m
e[32mApr 17 18:26:02 GobanSocket.ts:289 GobanSocket reconnecting in 50mse[39m
e[31me[1mApr 17 18:26:02 ! GobanSocket.ts:206 GobanSocket error unable to verify the first certificatee[22me[39m
e[33mApr 17 18:26:02 GobanSocket.ts:217 GobanSocket closed with code 1006: Abnormal Closuree[39m
e[33mApr 17 18:26:02 main.ts:143 Disconnected from servere[39m
e[32mApr 17 18:26:02 GobanSocket.ts:289 GobanSocket reconnecting in 258mse[39m
e[31me[1mApr 17 18:26:03 ! GobanSocket.ts:206 GobanSocket error unable to verify the first certificatee[22me[39m
e[33mApr 17 18:26:03 GobanSocket.ts:217 GobanSocket closed with code 1006: Abnormal Closuree[39m
e[33mApr 17 18:26:03 main.ts:143 Disconnected from servere[39m
I’m not sure what causes this, there do seem to be other bots online.
And on the note of support: Since the beginning of this year, reports on cheated bot games (low ranked players wining vs 9d+ bots, messing up the ranking system, or even worse, people abusing the auto-score system) are systematically ignored. Is the reporting system broken, or is cheating now allowed vs bots?
2 type of reports. The first type is like this: prapc2555 vs. 20bTurboLz-Elf-v1 (R471, i juist made a report)
that i report as “cheating at scoring”, that i find most problematic since it’s not even somebody using a bot to win, but they flat out abuse the auto scoring system.
The second report is like this game 友谊赛 (R742) where a 5d wins against a 9d+ bot. Ok, sometimes these games are legitimate, and the mod team should have tools to detect that, but 99% of the time these bots should be super-human and there is no way somebody legitimate wins. (The only way to win vs these bots is to use their blind spots like cyclic groups, big dragons that the bot do not understand etc.). Nobody is going to slowly pull ahead in 120 moves vs these bots.
In the past, i would usually get a mod response of this player is warned, or this game wasn’t cheated. But since this year, my reports are just ignored, and i’m losing track of what game i’ve already reported and what games i’ve never even tried. There is now a backlog of over 4 months of games to report, might be close to 100 of them over all 4 bots. (well, one doesnt get reports since it’s only 5-6 dan so wins could be legitimate, even while i can see at least half the wins are probably cheaters).
One final note: The report system seems to limit reports to a maximum of 5 at the same time, so even reporting all 100 is a chore. I tried a few times, but quit each time after i had 5-10 reports ignored. I hope i’m not hitting some kind of report-spam protection?
Hah, it’s cool that basically all (I guess ranked, with no handicap?) wins against these bots can be ~automatically considered cheating. This sounds like quite a valuable signal for detecting cheaters, perhaps it should even be automated?
The first game was lost by the bot because the auto-score evaluate the game different than the bot. The bot sees a 42 point win and is willing to pass on that (check score estimate, is shows a similar score). However, auto-score merely marks some corner stones dead and doesn’t award mojo, so black wins falsely. People repeat this game many times and farm the bot for rank this way.
The second game the bot resigns at 1.4% winrate, score estimate at this point is almost 80 points because white captured some huge group.
Now it’s not said any win vs these bots is automatically cheating. As there are some weaknesses, however any win vs these bots do warrant a very close look, somebody that was losing vs 4k players 1 month ago is not winning vs a 9d+ bot today, the game itself also shows me no indication any known weaknesses of the bot was abused. These are all very strong indicators, and i sure hope the mods have even more indicators.
I’d say that’s not a false win for black. White passing in this position is a beginner mistake. Is that really how KataGo usually plays, or is this exploit possible because of some configuration issue?
Because of such exploits existing for most bots, I’m all for bot games being unranked, except perhaps for provisionally ranked players.
That’s my view. If you don’t like your bot getting beaten because it makes mistakes tpks learn to avoid, tough luck, bots don’t get a pass just because they’re bots. So the problem is so difficult it wasn’t even close to solved until 2 decades after chess? live with it
I assume this is referring to " (roleplaying games) Initialism of totalpartykill: the situation where all of the player characters in a role-playing game have been killed.