site banner

Friday Fun Thread for December 5, 2025

Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.

1
Jump in the discussion.

No email address required.

https://www.lumine-ai.org/

We introduce Lumine, a generalist AI agent trained within Genshin Impact that can perceive, reason, and act in real time, completing hours-long missions within complex 3D open-world environments.

Unlike most of these science papers they actually show video of it working, which should be applauded IMO. It's very strange watching the AI's chain of thought as it thinks about what it's doing on screen, I only watched a few minutes and it was doing that standard chain of thought thing 'hmm I shall completely misunderstand the situation, misidentify this random mob as Stormterror the boss dragon but I'll still do the right thing anyway and kill the mob.' Despite the weird thinking, it just works.

Genshin combat is fairly simple since you just press 1 e, 2 e, 3 e, 4 e and do some left clicking, maybe a few Qs and dodge the telegraphed attacks. But there are also a fair few puzzles and more complex activities that I'm sure should give Gary Marcus conniptions (freeze the water to get the anemoculus on it, accomplishing tasks over a long period without getting confused, generalization beyond the training materials into other mihoyo games). I would've liked to see it have a go at the more complex puzzles in Inazuma though. They're using a fairly small model and fairly small compute too by corporate standards at least, real-time AI video processing isn't going to run on your PC.

It also shows the paucity of 'time-horizon' measurements. 5 hours of Honkai Star Rail, where does that fit on the METR chart?

I watched ~10 minutes from the middle of the first video and this looks super fake. There's no way a small/fast model could seamlessly navigate a 3d space using wasd+mouse (which is extremely complicated compared to the controls available to a self driving car), and then it goes and does things like jumping over a wall that was visually obstructed without pausing for a second. It also switches between using shift and right click to dash, which makes absolutely no sense for an agent to do.

If they wanted to fake it, why go to all this effort making video that backs up their claims? Why would bytedance even try to fake it, they're a company the size of facebook, an AI leader, not even publicly traded, they have no need for prestige or investor hype (of which this has gotten roughly zero with its 224 views on youtube).

It's a model with over a thousand hours of training for exactly this kind of 3d space movement, in this very game, I see no reason why this shouldn't work. Claude's struggles with pokemon are because it's a general-purpose model, not a pokemon specialist.

Also I don't see what's wrong with the clip of video you linked. It looks like a character getting over a perfectly visible wall to me?

which makes absolutely no sense for an agent to do

Yeah no agents make sense. Sometimes they delete the D drive. Sometimes they have depressive meltdowns. Sometimes they want to take a break and look at pictures of the zoo. Switching between shift and right click is not a big worry.