Can ChatGPT Win a Game of Pokemon Showdown OU?

The age old question, can a weird little bot with all knowledge of the internet beat up a 5 year old in a children's video game?
Become a Patron: patreon.com/kingtheluck
Follow me on Twitter: kingtheluck

[@KingTheLuck]

I have been informed that I forgot to remind you to subscribe, and that people apparently don't subscribe unless you ask. So don't forget to subscribe lmao it helps me out more than you might think!

[@LilacMonarch]

missed opportunity to say "I have been reminded to remind you"

[@traylene1139]

I feel like chatgpt found out about togekiss+excadrill for doubles and then never wanted to think of anything else, it always suggests that from what ive seen, even for singles

[@Boricosa]

Dynamax and its consequences

[@arazet]

It’s not the ai’s fault excadrill is so cool

[@Mepphy99]

"your move is reasonable" "switch to Clefable" that killed me

[@appointmenteer]

this is a perfect encapsulation of the fact that gpt is just filling in words that it think might come next. it has no fucking clue what is going on, just that certain words make sense to go next. its so incompetent in a lot of situations and its hilarious to see here. good video

[@Chirrie]

people think it's so good at stuff but then experts at a topic'll see what chatgpt says and it's really not as good as people call it

[@lyingcorn]

Tell ChatGPT "when I was 6, my sister was half my age. Now I'm 70, how old is my sister?" There is a 50% chance they say 73

[@TheKyubiisaan]

@@lyingcornas somebody who just did this puzzle in professor Layton I don’t blame them.

[@niranekomata3383]

@@lyingcorn chatgpt is pretty dumb at math but I also heard we (the users) made it dumber over time and if that's true this is sad news.

[@lyingcorn]

@@niranekomata3383 it's memory cut off in 2021, so we didn't do shit

[@joelkilchoer4238]

With some of these sets I’m convinced that everyone else was doing the same challenge as you.

[@obiwancannoli1920]

Nah it's just low ladder

[@draikken_]

I wonder if ChatGPT didn't recognize that Ninetails-Alola has different typing than Ninetails, it would explain why it was insistent on using Scald.

[@erichluepke855]

Almost certainly that is what happened

[@the_mothdegenerate973]

i assumed it was because Scald is pretty well known to be a great if not broken move, and whatever data ChatGPT has access to would no doubt mention it being a good move repeatedly, so it assumed that Scald is a great move to spam in every situation

[@EdgarAllan2pointPoe]

​@@the_mothdegenerate973I have little doubt that when historians peel back and decode the exact version of ChatGPT's language model that was used to make this historic video they will find that it was actually a combination of both of your theories.

[@josephmother2659]

@@EdgarAllan2pointPoeyou are probably right lol

[@shxd0wwazhere]

True or because it has a chance of burning the target chat gpt might thought it's a fire type move

[@10001vader]

Considering that the zapdos had been spamming roost, going for the earthquake wasn't the most terrible choice. would've likely oneshot the zapdos after a roost

[@qwerty99846]

i can't tell if this is /j or /srs

[@alondite215]

@@qwerty99846 I mean, Roost _does_ cause it to lose its Flying type for the turn, making it weak to Earthquake, so it's not entirely unreasonable.

[@qwerty99846]

@@alondite215 i did not know that

[@leebulger7112]

​@@qwerty99846I guess when birds are roosting they are not flying.

[@LilacMonarch]

@@qwerty99846 yeah that's the big downside to using roost

[@smollmoth6376]

Am I the only one amazed that chatgpt potentially knew that roost grounded the pokemon and that's why it chose eq. It wasn't a smart decision but I didn't think it'd even know most pokemon nevertheless the effects of moves.

[@dropfish3109]

i think it just clicked randomly

[@dc_dachi]

I think it just saw electric type and said to click EQ without checking for immunity

[@dylanb2990]

@@dc_dachiwhy assume it understands weakness but not immunity?

[@ududusjjs]

​@dylanb2990 I mean pre gen 3 pokemon Battle AI did the same thing.

[@formerlypie8781]

You sas all the dumb shit the ai did and you think that it knew that roost removes flying type, which would allow earthquake to hit it? The bot didnt even know what mons it had on the team it made

[@corngo9468]

Had to really pause the video to try and comprehend the Gengar going from perish song into power-up punch

[@TabletPencil]

AI has always been able to play pokemon, just look at the trainers!

[@MrCheeze]

*uses agility for super effective damage on poison types*

[@Lordbowser666]

@@MrCheeze work like a charm, that move never failed me (gen 1 AI)

[@marcoasturias8520]

Sure, but can it win?

[@bujustic]

Trying so hard not to "well actually" this comment

[@Gatebreakerglitcher]

@@bujusticgood, don’t cause that’s cringe af

[@Phantanatomy]

8:11 This is literally just that one meme of Patrick and the wallet

[@a.v.y8331]

based off all the ChatGPT content that creators have been putting out, I think it's clear that ChatGPT is decent to pretty good at a surface-level understanding of anything you ask it, but if you try to dig deeper and get to an expert or professional level of understanding, it falls on its face hard.

[@monstermaster4524]

The thing with ChatGPT is that it's not the best AI, just very popular and that carries it.

[@cuchococh4977]

In defense of ChatGPT the thing couldn't see nor was informed about how much health mons had, I wouldn't be easy even for a human to play without that or other vital information like damage calcs

[@jjpendejo7407]

It used draco meteor on a clefable and earthquake on a zapdos

[@Tantrum2u]

@@jjpendejo7407chatGPT def thought it was pre gen 6 (clef would be normal) and that zapdos was gonna pick roost, easy mistake to make

[@blueyandicy]

That actually sounds like a fun challenge to be honest.

[@RaynmanPlays]

Nobody: ChatGPT: Switch to Clefable.

[@SPZ-gv2on]

Honestly that's a much safer swap into zeraora than drill is. Zeraora runs close combat 90% of the time and that + knock off would do a million to drill.

[@wombatpandaa9774]

I can see this being difficult for Chat GPT partly because it can't analyze video and that's where a lot of the good stuff is. But it would also help a ton if it remembered its own team comp

[@McC0n0r]

I did wonder if more information could have helped, like the status of each Pokemon, how much damage was being dealt, current HP, etc. Considering some of these plays, probably not but you never know. Also I just want to say that as a Volcarona enjoyer, that Volcarona VS Heatran play hurt me on a Individual Value level.

[@facelessmoon8105]

I've tried this and it actually makes it worse in my experience. The more input it has, the easier it is to forget stuff, so it loses track of the game state in around 10-20 moves

[@cinnasauria]

I remember generating some teams last year, and my favourite was one it described as "a fairly standard Prankster RestTalk Toxic stall team." It described all Dracovish's moves as being "for STAB" and said "this team does not have a sixth Pokemon." Intimidate (Grimmsnarl) @ Leftovers Ability: Prankster EVs: 252 HP / 252 Def / 4 SpD Impish Nature - Parting Shot - Rest - Sleep Talk - Toxic Grassy Surge (Ferrothorn) @ Leftovers Ability: Iron Barbs EVs: 252 HP / 88 Def / 168 SpD Relaxed Nature - Leech Seed - Protect - Spikes - Stealth Rock Merciless (Toxapex) @ Black Sludge Ability: Regenerator EVs: 252 HP / 120 Def / 136 SpD Bold Nature IVs: 0 Atk - Haze - Recover - Scald - Toxic Spikes Gale Wings (Corviknight) @ Assault Vest Ability: Pressure EVs: 252 HP / 4 Atk / 252 SpD Careful Nature - Roost - Brave Bird - Defog - U-Turn Haze (Dracovish) @ Choice Band Ability: Strong Jaw EVs: 252 Atk / 4 SpD / 252 Spe Adamant Nature - Ice Fang - Psychic Fangs - Aqua Tail - Crunch

[@aname4390]

With how much GPT likes draco meator, I think it would do much better in gen5 OU.

[@CodPatrol]

What’s good in gen 4?

[@kraykyle]

Wait, but can we talk about that insane read by Chat GPT at 7:45 predicting the Hawlucha switch in?

[@dynosgarcia7645]

Whatever happened to the actual AI made to play Pokemon showdown?

[@electricnezumi]

3:12 1035 Elo is like, one or two matches won. A lot of people on the OU ladder aren't that into competitive, you only really get people who are looking into the principles of teambuilding and the good strats and stuff at like 1200 Elo

[@imp4209]

Yay I'm 1250

[@imp4209]

And I run around with some of the weirdest strategies on pokemon showdown, my teams have like...1 or 2 ou mons most of the time and the rest are conscripted from uu and ru

[@BRW667]

Damn, chatgpt is better than me at teambuilding😅

[@user-uv2cp1qd1j]

Oh my god! I just clocked the username and realised I played you during this challenge. I think I beat you bc the team is obviously not very good, sorry!

[@servine1212]

Nah bro, deagon dance shadowball dragapult is the new meta lure set

[@UltraSharkBros]

13:55 I thought I was crazy and having mental visions of 50 ways to say Goodbye by Train 🤣

[@shrambles]

This "ChatGPT" individual is trifling 😂, dang they suck at pokemon 😂😂

[@frogtoad8692]

Can’t wait to see chat GPT in nat dex Ubers

[@MegaVirus700]

Honestly, though, I love low ladder because they're cooking. Half their pans are an oil fire, but they're cooking

[@pksprite6401]

Dude your little character guy is terrifying

[@gamez4420]

I'm impressed it had the forethought for hazard control

[@AshenDust_]

You losing to Regieleki when you had a Ground type is pretty funny. It can’t even hit you neutrally with Ancient Power or Explosion because Excadrill is part steel type.

[@SPZ-gv2on]

Drill still has no recovery and life orb recoil + rocks chip adds up. Most excadrill teams are quite weak to eleki in practice unless it's leftovers drill.

[@ieatwater6550]

​@@SPZ-gv2on He lost because he sacked drill to lando, nothing to do with chip. Ignoring that, your argument that drill teams are weak to eleki is extremely shaky. LO Drill is not a pokemon that you want to give free switches for and Eleki itself is incapable of dealing significant chip to it. Not to mention, drill is commonly run with other pokemons that check eleki, such as ttar, hippowdon, dragapult and/or ferrothorn.

[@AshenDust_]

@@SPZ-gv2on you’re letting Excadrill come on absolutely free unless you make a risky double every single time you get Eleki in, Drill should be smashing holes in teams before it dies to chip as long as you don’t sac it to Lando because you can’t be bothered to run calcs.

[@mailboy1940]

Was that 50 ways to die by train in the background at the end?

[@mx.benzedrine]

The Train instrumental at the end caught me by surprise. /pos

[@StormEagleCH]

The psych up Clefable had me rolling

[@Pidgeyato]

Is that 50 ways to say goodbye

[@ashblossomandjoyoussprung.9917]

Nasty Plot Poltergeist?

[@3cho621]

at 4:28 when you're facing a team with "normal human person sets" that's actually a bot on ladder...

[@FappleJackity]

I love microdosing shredded cheese

[@ohhi687itsme8]

This was actually suprising.

[@Eroil]

The last opponent was probably using ChatGPT to play too

[@Emulleator]

have you considered using Draco meteor?

[@jaksida300]

2:44 Verlisify sets

[@requiemagent3014]

Seeing those OU players explains why they banned so many pokemon in gen 9 OU to make another tier. This isn't a joke. Bunch of clowns

[@Hiushisan]

Early in the video you seemed to criticize bulk up mimikyu, but I've personally had success using that on it. The set of Bulk up, Play rough, Shadow sneak and Drain punch, holding Rocky helmet is pretty reliable.

[@SPZ-gv2on]

It's very bad, sadly mimikyu just doesn't have good stats or moves. Play rough and shadow claw are weak and it doesn't have the attack to make up for it. Disguise is only so good.

[@imp4209]

Wood hammer, return, knock off, drain punch.... all this good coverage going to waste

[@IsMyNameUnique]

for this to work u need to tell it the description of the field (who is out, what status effects each has, etc) alongside your moveset and the movesets/pokemon of your entire party alongside what the enemy's entire party is. every. single. turn. chatgpt forgets old messages quickly, it only holds onto the last few messages so u need to basically constantly give it an entire paper on whats happening ever turn theres a time restraint so that makes it almost impossible though...

[@CodPatrol]

One thing I wish pokemon did better was make moves do less damage, I’m tired of battles ending in one turn because a move is so powerful. It’s supposed to be a back and forth effort with trainers carefully picking their moves, not just a one and done sweep. Stall is incredibly good for the game, yeah some pokemon are ridiculously bulky but in general it’s good for strategy and game plans.

[@vishnubajpai4676]

Great vid., maybe u can just copy the effects of the round to the chat as soon as it happens and give me a pokepaste before the match starts to refresh him

[@Krispizz]

It’s almost as if this robot doesn’t know what it’s doing

[@LilacMonarch]

and it has very bad memory, so it forgets its own team

[@altariamotives16]

is that 50 ways to say goodbye in the bg?

[@DoggoDoesStuff1]

I love how you apologize for your rudeness at the end lol

[@Dave_Chrome]

10:32 what Pokemon battle music is this? I've heard it before in videos, never got a name for it though

[@KingTheLuck]

I think that's Nemona's battle theme!

[@Mut-ant]

I tried this directly after chatGPT release too and it kept using some kind of inverse type chart for its decicions XD

[@ivanpaijolin9128]

You are so underated bro. Keep it upp

[@DeadGirlCherry]

Can someone tell me the name of the song playing the final battle? I feel like I absolutely should recognize it but I can’t place it 😭

[@rocketmudkip]

NOOOO AI WILL TAKE OVER THE WORLD!!! AI:

[@djelibaebi]

one million vidoes of can we outsource the thing to the robot that pretends it knows about things and watch them do it a bit badly.

[@asterlofts1565]

Of course can... but ChatGPT need learn the game more time. Like the players.

[@eugloopydilemma]

Good to know that we still probably have at least a few years left before AI takes over the ladder, thanks for finding out.

[@DorthLous]

... Why 3.5 and not 4?

[@KingTheLuck]

3.5 is free

[@Pyrodiac]

@@KingTheLuck 4 Isn't? Huh, good to know.

[@ulrichbrodowsky5016]

At 7:45 you completely destroyed the game of ChatGPT. It was reading the switch to Hawlucha but you forced it to use shadow ball instead!

[@robplaysstuff6029]

10:22 chatgpt can beat fear

[@TR-kn3sn]

These people have elo! hovers over someone with 1035 elo. The ladder starts you at 1000. These people do not have elo.

[@Accrovideogames]

To be fair, you made a lot of grammatical and spelling errors that may have confused Chat GPT. For example, you typed "time" instead of "team".

[@KingTheLuck]

In my defence I had like 10 seconds before the time ran out most battles

[@CodPatrol]

Having a burn or toxic chip away at somebody’s health while the pokemon absorbs attacks with its high defence is a fun way to play the game. Do you use protect or double team? Protect or bite and hope for a flinch? These are the moments that are great

[@genib6174]

GPT4 and some prompt engineering could probably make a huge difference. This test is not serious!

[@ari_cool]

i went out of the livestream to watch this

[@ParasocialCatgirl]

ChatGPT is just like me fr

[@pumba6139]

how do u get that graph at 2:18

[@KingTheLuck]

Marilland type matchup analyzer I think!

[@Bronze998]

You could of won more if you used Draco Meteor more

[@naturelulI]

4:55 Yes! I agree so much :D Stall has always been my preferred style (sorry) because if I make one bad mistake and lose a key pokemon, its all over. It really puts my skills to the test. My favourite metagame was Gen8 AG largely because I loved the stall teams made out of the universal Necrozma-Dusk-Mane, Eternatus, and Yveltal core. Well except Necrozma-Dusk-Mane because I hate that thing with a passion XD. Of course I don't only play stall. In Gen8 AG I challenged myself to get 1800+ rating with a stall team, balanced team, and hyper offense team. And in Gen9 I've been avoiding stall because I feel like the terastalising mechanic is better for offense than defence. I wish I could still play Gen8 AG...

[@crarlo]

stall is really good right now tho, this might be your time

[@mattskord9178]

Noticed the siivagunner

[@OliverSpencer-sd9ou]

time stamp?

[@sebastiaan3265]

Very based and epic and cool zelda ds games background music

[@meso07]

do the same but with the gpt 4.0 version, it would be really interesting to see if that would make any kind of difference you could even have it read over a guide or something before to see if that would also make a difference

[@baileyayyy5085]

at low elo you can probably click random buttons and get a win within 100 games lol

[@catsanzsh]

Next time use openai playground or a a custom lm ;3

[@AesopFable]

What was the song after 12:58?

[@shadyspiritomb5285]

Kind of sounds like the donkey kong country soundtrack

[@honjon666]

Gangplank Galleon. Idk which iteration, but doesn't sound like the snes arrangement

[@jaykay8464]

Siivagunner's Gangplank Galleon rip: kzbin.info/www/bejne/o5vbgISLd92soqs&ab_channel=SiIvaGunner

[@frogtoad8692]

Comment for algorithm :)

[@KanderUdon]

Nice video

[@Mudcrown]

This is so cool

[@JustKosmik]

Eoe it ssid no views

[@spit-fire]

I hate to break it to you but 1200s is not a respectable rating 😂

[@KingTheLuck]

Half way to top 500 with this shit ass team ain't bad in my books lol

[@dancingllama2392]

I'm better then ai at pokemon. yay

[@iamlorddems3859]

Remember chatgpt was trained to chat not to play pokemon

[@-_soy_-888]

Why are you going over time outs? You're just wasting time

[@WannabeMarysue]

first

[@MozzarellaAndChedda]

The real question is if i can win showdown ou 🥲