I made Jev play Flappy Bird against 3 other models. Who do you think won?

i wonder how much of this is the latency of the llm response vs jev

a decent amount

How do you get Jev to play these games? How do you give it the game state?

Thanks! Can you imagine what Jev would be like with eyes?

what state are you giving jev?

No pixels: Jev gets a small JSON of the bird's position and velocity, the next two pipe gaps, and a bunch of options on when to click flap and it can choose when, in seconds.

The same thing is sent to the other models. I could send them images but they would be way too slow so they just lose the game.

How it answers so fast! I think in future online multiplayer video games will be obsolete due to AI

jev!

My name Jev

Jeff

What was the cost difference?

Jev was 3x to 18x cheaper than the alternatives.

Do it again but have the LLMs write a script to play the game based on the game state.

They respond based on game state right now anyway. I tried to make them use BU or CU and they just couldn't respond fast enough.

jev and clash royal 🧨🧨

I think you just came up with the first benchmark for Jev. The Flappy Bench.

Haiku actually did better than I thought it would

Me too lol

GLiNER can do computer/browser use like Jev while being local, free and open weights. it's so lightweight it'll run on a laptop.

Full video comparison here 👇

bro this is pretty stupid. the model is choosing between hitting something or not. we already now its fast too. so comparing it with the text generators makes no sense.

Бля, пожалуйста, хватит давать Jev делать хуйню, придумайте уже прикладное полезное использование…

Ok, I did, back tested it on 850+ support tickets for my company. Watch this video here.

Bro, gpt 5.4nano? Really? At least make it compete against 5.6 Luna lol.

Luna would not respond fast enough to play the game... LOL

i am also testing it out, how do you think it's doing the task so far? any good ?

Which task: playing Flappy Bird? Yeah it's pretty good at that for sure.

that i know, but doing multiple tasks at once, like agents

Opensource it!

The code was one prompt