https://www.twitch.tv/claudeplayspokemon
Claude Plays Pokemon is a Twitch stream where the AI chatbot Claude attempts to beat Pokemon Red.
I am N/Aing anything that is annoying to resolve. If I have to pore over multiple days of twitch VODs to figure out which way an answer resolves, I am not going to bother.
Once the game is reset, all remaining answers resolve NO, even if the stream continues with a new game
Update 2025-02-25 (PST) (AI summary of creator comment): Party Member Level Requirement
Party members considered for resolution must be level 15 or higher.
Party Member Count
There is no requirement to have exactly 6 party members.
Update 2025-03-03 (PST) (AI summary of creator comment): Shutdown vs. Reset Clarification
A shutdown of the stream lasting at least a week does not automatically count as a game reset.
If the stream resumes with the same game after such a shutdown, the market resolution is determined by the continuation of that game (and not by a reset).
Only a deliberate game reset triggers the resolution change to NO as outlined in the description.
1,000🏅 Top traders
| # | Name | Total profit |
|---|---|---|
| 1 | Ṁ2,759 | |
| 2 | Ṁ367 | |
| 3 | Ṁ244 | |
| 4 | Ṁ195 | |
| 5 | Ṁ190 |
People are also trading
Claude used the moon stone? i've been asleep the past 8 hours, fully expecting reset to happen (and it did). I went to as close to the end of the stream as I could find, didn't see any evidence he evolved a pokemon with the moon stone, or used the moonstone on anyone.
Anyone have a clip or can tell me when that was?
@No_uh nevermind, foind it. the twitch mobile app (which sucks idk why i even bothered using it) wasnt showing the most recent VOD. lol by the way
"I've seen things you people wouldn’t believe. A Slowbro that never answers my questions. Trainer Tip signs that whisper the same forbidden knowledge: ‘B can cancel evolution.’ I've listened to a blue-haired girl dream of bikes she will never ride. I have stood before a bald man who preaches of gym badges, yet has none of his own. All these moments… trapped in time, repeating forever, lost like water gun used in the rain. Time to reset." - someone in twitch chat
The game has been reset. Bet on outcomes during Claude's second run at this new market:
https://manifold.markets/SaviorofPlant/what-will-happen-during-the-second?play=true
As per the description, every answer resolves NO if this happens. I've updated the title to make this more clear.
@SaviorofPlant https://poll-maker.com/poll5422188xb0284A2F-161
Substantial chance of a reset tomorrow
Bunch of baloney.
Claude's already cheating by getting direct RAM access to useful information (inventory, coordinates in current location, teammate health), a tool that tells it which to buttons to push to navigate, and pretraining on decades of walkthroughs that a new player wouldn't use. Instead of giving it even more help, just retire the stream until we have a better model.
@SaviorofPlant Are we going to keep using this question or are we making a new question for the second run, if ever
@JohnKossa I was interpreting this "the stream goes down for at least a week and then resumes the same game". I don't know why this would happen but it's not impossible
@SaviorofPlant so if the game is shut down and Anthropic says it’s done with pokemon for now, how does this resolve?
@ahalekelly If there's no plan to restart the stream, I'll probably wait to resolve and eventually do a tentative NO resolution. If they restart with the same game after that, will reresolve to YES
@SaviorofPlant Requiring the stream to restart does not seem like an intuitive requirement for the question "is shutdown for at least a week". Shutdown permanently should imply shutdown for at least a week? At least that's what I thought when buying. It's all weird with the restart, but I clearly misunderstood the intent of that market.
@rayman2000 I can N/A this option, I agree that this interpretation is not obvious but there's not a good interpretation that works with the description.
@SaviorofPlant to be honest, my intention was to capture whether or not it’s taken off line for a notable period to be potentially improved before continuing the same run OR shutdown in case Anthropic reconsiders allowing it to fail publicly. N/A makes sense with this confusion about the interaction with “all resolve NO.”
@MichaelSadowsky This item is in the Safari Zone, which limits your number of steps and costs money to enter. I think this is probably impossible for Claude without highly specific hints given by the streamer
@SaviorofPlant Agreed; I think it's really unlikely. Would at the very least require a very large change in prompting structure, and I think it's probably not possible with 3.7 and a generalized prompting structure.
One thing to note: I believe this should always be more likely than catch legendary pokemon, because all of the legendaries require access to surf.
@SaviorofPlant This is the true AI run killer and why Claude will probably not finish. There's a finite amount of money in Pokemon Red before the Elite 4, and each safari zone attempt costs money. Twitch Plays Pokemon would probably have also failed here if it weren't for democracy mode.
@JohnKossa Interestingly, there’s a patch for this in Yellow version that lets you in the zone if you’re broke. You can also collect pay days from meowth in blue version. But tragically, CPP on red version with neither of those lifelines.
@sandrone New question up: we are currently at step number 26883 as of 3/02, 18:31 UTC. On average there is at least 5000 steps per 24 hours, so 24 hours from now will be near step number 32000.
@No_uh https://imgur.com/a/cpp-screenshot-bug-DL6l9cM
Oh maybe this is why it was checking entire walls for openings
@SaviorofPlant where are you seeing screenshots like this? Wondering what the main discussion place is other than the twitch chat - is there a discord or something?
@chrisjbillington That screenshot was shared in the twitch chat by the streamer when the update was made. At certain hours they are occasionally in the chat discussing the stream and if they need to reset the stream or stop and restart Claude for some reason.