aaaannd it plays Doom now. On device, 150ms latency, on 1B param model

Took me a few hours to get the training recipe all good. It’s not perfect but the above demo was a model fine tuned with only 7-samples in under 4 minutes on my laptop.

thats not doom bro. its slop

Yeah wtever proves the point lol

the whole thing with X can run/play doom is that it plays doom. would be cool to see it actually run doom

You really should've just had it play doom lmao. The people here a totally hung up on the fact you used a doom clone instead of doom.

We have doom (raycaster) at home ! 🤣

New CSGO cheats just dropped

the future of ai cheats

I’m going to use this.

Testing your Qwen open source RLCD version on Macbook pro M5 atm - will report back to you later if you like.

I care less about Doom than the 150 ms. If it stays responsive when the network is gone and the device is hot, that is useful on device AI.

But can I run it on an ESP32 chip yet? 😃

Hopefully this convinces gaming companies to let go of kernel level anti cheat. ^^

Very impressive!

that is NOT doom

this ain’t doom

Love to see a cost comparison; same game. To see the cost savings...

Rough napkin math, this costs roughly 20x less at least

I'd read video feed,for 24 hrs is $4300 for before this. And $7 for this for 24 hrs. I'm not sure if that's apples to apples or not. But that's a 614x diff. @rehan_shei on the 30th had streamed a Rick and Morty cartoon that the chat was able to bend interactively. Really cool idea. Copyright lawyers asked him to take it down. He did. But this showed a very cool idea. The expense was to high w/o heavy advertising and cost. But this seems to radically alter that cost. This is why im asking. Thx if you could help here. And ty for doing your testing towards this period. We're living in extremely interesting exponential times!

"- $4,320 / day = H3 Max 480p **video-seconds** at $0.05/sec continuous. - $7 / **hour** = TypeSafe Doom-bot **decisions** at ~10 Jev calls/sec. Corrected combined room: Jev cuts the director bill ~40–95× vs a frontier LLM. Video is still the expensive meter unless you reuse beats or drive a realtime renderer."

Future was always for S(mall)LM

Damn lol

lol next?

Does it play Terraria and League Of Legends though

150ms matters less than p99 latency once schemas get deeply nested

how is it perfectly tracking enemies if it has 150ms latency? is the game slowed down?

1B?!?!?!

Backrooms final boss

I'd rather play games myself, do you have it integrated with browser control or desktop control? In any case great start.

Really impressive. Does it still have CoT under the hood?

1B model playing doom is wild

what model did you use to build the game?

It doesnt play doom, it slays doom

Can you do an intelligence comparison?

every new tech eventually has to run doom. 150ms is like 6 decisions a sec, more apm than most people playing it lol

Minecraft next