Claude Mythos generates a fully functional, multiplayer Minecraft clone using zero-shot coding

It’s funny to see how little ambition people have when testing new models.

@test_tm7873 what

@mattshumer_ most people just wanna know if it can make memes and write their emails faster

@scaling01 Z ai or Z ai ?

@mattshumer_ Did I read earlier today they need to slow the Ai down now since it’s getting a mind of its own? Scary bro https://www.youtube.com/watch?v=U3kOzeovweE

@scaling01 There is zai lab and zai discord.

@scaling01 the glm guys??

@scaling01 voxel terrain, chunk generation progress bar, working hotbar - that's not a toy demo. multiplayer on top of that in one pass is the part that should make every competitor nervous

@mattshumer_ ppl are using Xbox svg as a judge of mythos when gemini2.5 was matching it.. doesn’t that kinda prove it’s a dumb benchmark? Don’t really get the logic.

They can just train on these repetitive tests

@scaling01 If I were to play devils advocate… MC is probably the most written about, mimicked, and modded games in existence. Could there be more MC/related code in the ether than anything else? Noise augmented plagiarism?

@scaling01 mythos output looks exactly as insane as expected, you sound big mad for no reason

@mattshumer_ most people are still running hello world prompts on gpt-4 level models. meanwhile the real test is whether it can handle 6 context switches in a single agent workflow without hallucinating your database schema

@scaling01 people said the same about o1 and we all know what happened

@mattshumer_ Should they use models to release reflection and claim fake benchmarks instead?

@scaling01 The sentiment shift in AI moves faster than the tech itself

one impressive output and suddenly everything else feels outdated interesting how comparison has become the main way people evaluate value now

@scaling01 @test_tm7873 Zai the discord or the AI lab?

@scaling01 What’s insane about it? We have seen some pretty good Minecraft-ish results no?

@mattshumer_ "i saw you leave a file under "gradient descent" once" tier sentence structure is crazy

@scaling01 why would you believe that to be mythos lol, highly doubt anyone can put out mythos output out there right now due to NDAs they signed to get early access or seats

@mattshumer_ fewer hours in the day these days