Did Chess.com Upgrade the Bots???

Sort:
redapplesonly
Kratuu wrote:

Yep, Jimmy (600 adaptive) is now giving me more trouble than Emir (1000).

Yep, same.  Jimmy seems smarter, tho

BabYagun

@kemmrich, thank you for the examples. Do you think Emir plays weaker than a 1000 rated human should play? Or maybe Emir's rating is realistic, just he blunders queens too often?

About these Emir games examples, I would say: "He is ready to exchange his queen for a less valuable piece". Those are not pure giveaway queens.

BabYagun

@mistmade23, thank you for an example.

> To be honest the 1600 bot was even more challenging. 

There are 3 1600 bots, they have different opening books and settings. Which of them is more challenging than Li-bot?

kemmrich
BabYagun wrote:

@kemmrich, thank you for the examples. Do you think Emir plays weaker than a 1000 rated human should play? Or maybe Emir's rating is realistic, just he blunders queens too often?

About these Emir games examples, I would say: "He is ready to exchange his queen for a less valuable piece". Those are not pure giveaway queens.

 

I have only played bots so I don't know about "real" ratings.   ... and yes, in playing the bots I have won a queen early only to see my myself surrounded by minor pieces and lose all advantages.   

P.S.  Any chance you can have the bots play rated games????   That way the argument of a bot that is stronger or weaker than their rating would go away.

kemmrich
redapplesonly wrote:
Kratuu wrote:

Yep, Jimmy (600 adaptive) is now giving me more trouble than Emir (1000).

Yep, same.  Jimmy seems smarter, tho

I just played Jimmy a few times, 1 loss, 1 draw and 2 wins.  In both my wins I was over 90 in accuracy and in my first win, Jimmy was over 90 also (that is my first game against a bot where both of us were over 90!).  He has definitely gotten better!!

magicjohnson1971

Whenever I think something on the site has changed, I checked these forums and bang, here it is. 

Feels like I instantly got worse overnight...glad to see it's not really true but it still feels like it... 

I usually play Antonio & Nelson and both are much harder now.

mistmade23
BabYagun hat geschrieben:

@mistmade23, thank you for an example.

> To be honest the 1600 bot was even more challenging. 

There are 3 1600 bots, they have different opening books and settings. Which of them is more challenging than Li-bot?

Played Isabell. I needed a few attempts to beat that bot, whereas every bot up to 2000 was beatable in one attempt. On the other hand i replayed "her" yesterday without any problem. Of course i am not sure how accurate the bots rating are. For example this is a 1100 bot: 

https://www.chess.com/game/computer/11595671  https://www.chess.com/game/computer/11596355

In my opinion that was far to easy. Could be done in a few minutes. They all tend to blunder a lot, espacially queens are easy to capture.  

Here´s li-bot again wink.png 

https://www.chess.com/analysis/game/computer/11598327?tab=analysis

BabYagun

> Any chance you can have the bots play rated games????   That way the argument of a bot that is stronger or weaker than their rating would go away.

At the moment bots live in Vs Computer. To play rated games vs humans they need to be migrated to Play. That may happen in future.

Currently we have a place where bots play rated games vs bots. We know their relative rating. But humans feel the rating in a different way. Thus, we adjust bots skills according to feedback.

kemmrich

Still battling my way through.  I beat the first 1600 bot (Pablo) 2 out of 3, but I still had too many mistakes/blunders to be happy with him.   I'll try again later.  I'd like to be above 85 accuracy, or at least 80+ without numerous blunders!

I wonder if Pierre (1500) has been upgraded.  He usually plays the Bird opening with white and I thought I won fairly easily against that.   Now I seem to be struggling and it is a crap shoot if I can beat him.   To me, all the bots seem harder!!!

TheMsquare

I've so far selected the Sakura bot as a future training bot. Once I've beat em all. Those holiday mom bots were effectively all 2000+ bots.. maybe 1800 for the first one

BabYagun

> I wonder if Pierre (1500) has been upgraded.  He usually plays the Bird opening with white and I thought I won fairly easily against that.

Thank you again for your feedback. As I promised happy.png, I won't tell which bots were updated, to keep feedback bias free.

What I can tell: Opening books were not changed. If a bot played Bird opening with some probability , it will continue choosing that opening with the same probability.

A few bots became weaker this Monday, we lowered their skills based on user feedback.

Duck

*Bots move to /play*

 

"Let the farming begin!" grin.png 

kemmrich
BabYagun wrote:

What I can tell: Opening books were not changed. If a bot played Bird opening with some probability , it will continue choosing that opening with the same probability.

A few bots became weaker this Monday, we lowered their skills based on user feedback.

 

Pierre is still playing the Bird opening, just seems better than before.  But it is hard to tell, I can have a bad stretch against any of the bots!!  Jimmy seems easier today!

hypernova84

I don't know how relevant this thread is, but I thought I'd add something here anyway; I personally don't like the way a number of bots play. They had much more personality before, and to be honest I had based my progress and skill level around how I played against them. I thought I was between 1500 and 1600, would expect to win the vast majority of games against Antonio, have a tough battle against Isabel, defeat the odd 1800 rated bot. Now, I'm lucky to win a game against Antonio. That happened seemingly overnight.  Even Mateo, he simply does not make a mistake any more. Every move is just a little bit better than mine. There are no holes in his play, at least that are obvious to an intermediate chess player like me. Before, a bot may be attacking you with a clear tactical threat, but would hang a pawn in the process (just like a human), but now? No holes. If I set up an Xray or threaten a fork, he sees it. It just really feels like I'm just playing Komodo now. I get squeezed and squeezed, the position is equal, then it slowly gets worse. From a 1400 rating? Really? That kind of play? No mistakes, no bad moves, sees everything? I mean he's not hitting me with engine top rated moves, magnus style, but still.   

 

In my experience, since the update the intermediate to advanced bots play in a dramatically different fashion. They are much stronger. If you are going to make the bots stronger, then you need to adjust the rating, or people like me who play the bots exclusively, are going to get discouraged. I stopped playing chess for a while because I thought my progress had gone backwards. To be honest, the only reason I bought a premium membership to chess.com was because of the bots; they played like humans, they used an opening book (so I could get experience playing different openings), and I got very used to how each bot played. I could play chess like I play my other computer games, no rating, no stress. Now, the ones I play feel exactly like an engine. I can go and play stockfish on lichess for free.   

 

If the ratings were wrong in the first place then I have to ask how you guys allocated ratings. Did you actually have the bots play against 1500 rated players to see how they went?   

Martin_Stahl

@hyoernova84, do you have specific saved games with bots you can point to? Not all the bots have had changes, so staff are asking above to see examples to see if it's a bot that had been adjusted or if it's something else.

Stalkghoti

It seems like there was another update in the last few days.

 

I used to easily beat the 1300 bots, and am now struggling to beat 1000 rates Hans, who now plays much more aggressively and coordinated his pieces in ways even higher rates bots didnt before.

 

jewelmind

When posting about bot games probably a good idea to say whether they are timed or untimed games. I'm rated 1238, and I've beaten a couple of 1400 players in rated 15/10, but lost to plenty of lower rated players. I can say I find it easier to beat 1200-1400 bots, which is true, but I'm playing unrated untimed games for study, to improve my calculating and positional skills. These games are just not comparable to rated 15/10 against humans or bots.

AlexiZalman

Well, I am a low intermediate player who has been working my way up the bots - starting at the lowest - for the last three months. I play with infinite time as I am aiming for improved decision making. My subjective experience is as follows:-

(a) Below rating 1000 there was little evidence of any rating difference, the gameplay felt more or less the same. Play solid and you could be certain of a piece would be mindlessly, without pressure, thrown your way - usually a Queen for some reason. The openings were often very poor or very quickly went off-book and often quite silly. I didn't feel these games were of much worth.

(b) At some point between 1200 and 1400 there was a very marked change. Whether this was due to normal bot rating increase or recent changes have no idea, but very marked.

(c) At 1400 I now have a tough game. Openings are far more solid and to gain any chance of an advantage requires at least a serious position error or tempo losing move or a 2 to 3 move combination being present. You have to work at maintaining pressure on the bots and even with an advantage play is still very tricky with defence being much better and often surprising. You definitely feel any serious mistake would be jumped on.   Overall, I would say the 1400 bots are playing more of a classical than blitz game. This assessment is primary based on my recent last five games against Mateo(1400) who has accuracy around 80%, whereas mine has been 90+ with infinite time usage, position evaluation always available, and the odd takeback. Post-game analysis (off-line) indicates Mateo's unforced errors are fairly minor and that good play is definitely required. 

I am very happy with the gameplay, it's tough enough to force me to think and hopefully improve middle game decision making without feeling a win is a hopeless endeavour. 

Based on my experience I would scrap the sub-1000 bots as there is too little difference in gameplay between them. I am also inclined to suggest doing away with the ratings entirely and just have a range of personalities leaving the players to determine a suitable opponent. Alternatively, instead of trying to mimic low-level gameplay - so far badly - why not just allow the lower-level players to set a material advantage at the start of a game or do this automatically? Anything is better than a have-a-queen approach which fast becomes very predictable!

BabYagun

Thanks a lot for the detailed feedback.

> I am a low intermediate player

Does this mean 1500-1600?

AlexiZalman
BabYagun wrote:

Thanks a lot for the detailed feedback.

> I am a low intermediate player

Does this mean 1500-1600?

Using chess.com measures. I would consider 1200-1400 to be lower level intermediate.

Mateo is playing at a 1400-1600 level in my opinion, playing solid out of the opening, making rare semi-pointless tempo losing moves and is fast to capitalise on combinational errors, a bit weaker in the endgame. I have to put significant pressure on it in the middlegame for it to make a serious error but defends well afterwards - so making one good move is insufficient to win, you have to follow up the consistently good play otherwise you go backwards. My guess is that he would be quite lethal to a ~1500 Blitz player.  Note, I didn't play an aggressive/sharp opening which can often increase the playing strength of the bots.