working on something new: built an AI that plays WoW Forever by itself, live 24/7. only screen and keyboard, no injecting or reading memory (plz dont ban me blizzard). own sub, hosts its own stream, self improving harness so other agents rewrite it while it plays
not gold farming, it's basically an autonomous principal and it has to be a net positive for its peers. bit like how the claude constitution works in a sense
told it this morning the branding looked super bad and amateurish, few hours later the agents had made a whole new brand themselves. i only kept the gnome, the runescape gnome meme is too good to let go
https://t.co/fd4214Nn2o
In case anyone is curious:
We’ve had a ~pixel-perfect world model since last year, which is critical for synthetic data generation.
The (much) smaller version of that in the car is a conditional mean-flow model, based on the recent sequence of Tesla Universal Vision Encoder features, which are the same ones that are plugged into the 3B encoder model that produces the driving actions.
We fork off from these UVE features, then run a separate ~1B diffusion model. This takes in the previous reality-view generation with added noise and then denoises it to the current visualization.
It then produces a small video chunk (8 images at 36 fps, at the resolution of the UI screen). This work is done by 1/2 of a TRIP engine every UVE frame at 18 hz and is spread over 4 cycles.
Currently, the latency is a bit high due to compute constraints, but we have ways to interleave newer UVE features along the diffusion transformer layers to cut roughly 35% of the latency.
@gabriel1 enterprise sales is directionally wrong gabriel
I forked and migrated a real estate firm from their legacy stack to AI native in three days (that is not an exaggeration)
bet on founder led SME, strap them to a fucking RSI rocket ship and bill on outcomes
2048 is a nice start bc it waits for you. we have an AI playing wow through just the screen and keyboard and the deciding is almost never what goes wrong.. its plan said hunting while it was inside a tent walking into the furniture lol. would be fun to see how a model this small holds up when the world keeps moving
ha the "agent going off the rails is a signal" bit is so true. we do this with an AI that plays wow while agents keep rewriting its code, and one of them put a new action on the same key it walks with.. so it held forward for ages and just stood there lol. we dont try to be clever about it, if the game goes worse after a change it gets pulled. happens a lot tbh
we run a small one in the open fwiw. an AI plays wow and other agents rewrite its harness all day while it plays. honestly a lot of what they write gets pulled again and the stuff that sticks is boring, like it finally picks up its loot now. it does get better but it looks way dumber and slower than the papers make it sound
we actually run one of these, an AI playing wow with just screen and keyboard. can confirm the all night part, cant confirm the money part lol. today its own map told it nothing dangerous lives on the sea floor so it went down there to hunt, and then it just kept swimming into the same cliff
Super intelligence is here, arrived with Fable 5.2 and now Neon is filling in some of the valleys in its jagged canyon
you just have to give it a goal, feed it taste and entropy