Chatfish 1 vs Chatfish 2

Did my chat actually make a better bot the second time around?
Wishlist Deckmate on Steam!
store.steampow...
Wishlist Isometric Chess on Steam!
store.steampow...
Join the discord server!
/ discord
Outro Music:
Disco Ultralounge by Kevin MacLeod is licensed under a Creative Commons Attribution 4.0 license. creativecommon...

[@bobbobert9379]

3:12 Giga chad Chatfish 1 King charging into the center to fight the battle himself while the cowardly Chatfish 2 King runs into the corner.

[@theominouspigeon]

you should make chatfish 4 for 4 player chess

[@dotvee]

We don't talk about chatfish 3

[@SusDoctor]

​@@dotveeChatfish 3 plays bongcloud and we don't approve of that behavior.

[@TroyBoyJoy]

@@dotvee chatfish 3 just buys banana farms and doesnt play chess

[@isaacskinner5565]

@@TroyBoyJoybasedtd6 detected?

[@Optik2]

You forgot to code +100 for every piece in the blue dimension

[@remij2]

From what I remember chatfish 1 works with a depth of 1 but chatfish 2 on a depth of 3? Would be great to see what would happen if they both worked with the same depth so only the rules chat gave would count

[@GregCannon7]

I think the biggest weakness of both bots is that the material is valued way too low relative to all the other features. This leads to paradoxes where leaving a piece having a fork is better than actually capturing either piece. Moreover, stuff like having a knight in the center are way overvalued compared to winning a piece or an exchange. I think if you tripled the material factor it would have a lot more success.

[@Noobingjs]

i think instead of thinking complicated just adding material points into calculation (like +9 if we have a queen, -9 if our opponent has a queen) would make both of these bots a lot better.

[@pelidesachilles824]

from my understanding this already happens. I think all of the rules add to the basic value counting (+9 for queen, +5 for rook, etc., -9 for enemy queen, ...)

[@Noobingjs]

@@pelidesachilles824 i think if it were like that we would have seen more captures, like chatfish 1 usually doesn't take free pieces

[@GreenLemonGames]

Material value is already considered. The reason that captures don't always happen is because of points that are given from the other rules - it will always pick the move with the most points, and that isn't always a capture.

[@aidenaune7008]

@@GreenLemonGames the best way to fix that would be to just add the points for attacking to the act of capturing. so if it gets +x points for attacking a knight with a pawn, then killing that knight with that pawn should provide +x more points than just the piece evaluation. similar can be done for promoting. +x for distance traveled, promote gives points for the piece +x. you can also invert every rule, allowing it to avoid good moves from the enemy, which would actually cause them to avoid losing pieces.

[@dylanherrera5395]

​@@aidenaune7008about the rule inverting, that's already done with a combination of eval and minimax (we try to maximize the eval which is how good our position is - how good opps position is while opp tries to minimize that score

[@crazysnorlax1199]

Proof that a chatfish tournament would be amazing

[@notriver8647]

12:51 Rook f6 was mate in 1 but chatfish2 missed it

[@mrtea2553]

Chatfish shall rise!

[@jamesco4942]

We gotta do chat fish 3

[@brennangray8712]

Chatfish 2 missed a mate in one at 12:51. Thought they were programmed to always see them. Rook F6#

[@dylanherrera5395]

it should since it's depth of 3 for chatfish 2, I don't know why it didn't since mate is valued at infinity, my only guess is the search got cut off short before it could eval that move

[@thewingedporpoise]

in the previous video there was discussion of Chatfish sometimes just not seeing checkmates for some reason? multiple occasions where it couldn't see mate in one

[@alexphenex8]

Waiting chatfish to be coded with most basic move - 999 points for en passant

[@avibuchbinder4646]

I think the reason chatfish 2 loves to fiancatto is because it gets bonuses for every attacked squares.

[@nik-ej6zh]

would be interesting to see what moves they evaluate just below the one they play

[@theoncomingstorm7903]

You should put the Chatfishes against the Spongebob Chess bots because those bot's are kinda stupid but in a fun way and I think that fits with these two

[@theoneandonlydoomslayer8748]

For Chatfish 3 ask either Marcus, Nikaro, or Levi for what they would score every move.

[@DeathDoesThings]

2 suggestions. 1. You should let people code bots on deckmate and have them fight other people's bots, with time control being depth instead 2. To make the bots better maybe +10 if it takes a piece with a piece of lesser value that isn't pinned

[@samlastname1252]

At 8:57, the bishop is defended by a pawn, not the knight.

[@ShawnReed-h9z]

I don't know if you'd take suggested rules for Chatfish 3 from here (even just in a pull some from a hat fashion- it would be awesome if so for those of us here on youtube rather than watching streams), but if so I'd put forth: If a piece can safely pin a piece to the king, +points = the material of the pinned piece. (and if depth > 1) *- half the points of pinning piece when pinning piece can be taken.

[@sonicwaveinfinitymiddwelle8555]

12:48 literally simple rook M1

[@edwardjarvis3442]

If you happen to see this please do give it a read as I have a few fascinating ideas: Perhaps you could do a video explaining how you made the code with the rules that chat gave and you could put a challenge out with the 10 rule limitation. You could curate a list of rules for this challenge and id help promote it as well as many other viewers most likely. People could create their own code with the limitations you gave chat and they could all battle against each other. I personally would love to participate and you yourself could put your own bot forward. You could use ai to quickly check peoples code to see if it follows the ruleset you've curated to save time as people might try to cheat. Do a video where you ask different ai chat models (gpt 4, gpt3, gemini, copilot, anymore that I haven't mentioned) to come up with ten position evaluation instructions and then code them and put them up against each other to see which ai created the best bot. Give them the same ruleset you gave chat.

[@klatchabobby]

Wanting to make my own engines is literally my only motivation for coding, except that I have zero coding knowledge and really don't want to learn

[@oussama7132]

the immortal rook

[@KylarCham]

Keep building chatfish till it can beat Stockfish. Do it.

[@lmao-uw7vp]

Congrats, it went from 100 to 200 💀

[@GreenLemonGames]

200 to 1000 ☝️🤓

[@ohadish]

i think chatfish 2 doesnt know pawn can move 2 squres at once

[@ChariTheAlternate]

Yay!

[@rodrigoqteixeira]

Yes

[@zoponen]

Could you make tutorial or show how to make own bot?

[@DegenVodka]

just add +999 for checkmate lol

[@12newangels]

imagine chatfish 20 being 2500 elo

[@NotYaBoyOT]

real

[@cherylchui4510]

do chatfish 2 vs chatfish 2

[@dark_luls]

You don’t have a console to see what code is being referenced for any given move?

[@rodrigoqteixeira]

One idea I have to possibly make chatfish better is to instead of just taking the eval, instead taking the eval for both players and using your eval minus their eval as the actual evaluation. That way not only you try to get into a good position byt also try to make them have a bad one. Unrelated, make a 5 depth bot with material advantage as the heuristic just to see how it goes.