Since Opus 4.8 I've been formally verifying UI/layout in my custom graphics/game engine in Rust It has saved me so much time The agent writes the UI layout in a pure function (it's just math) which is invoked when rendering or writing proofs Then it can write proofs that hold for EVERY screen dimension. For example, "the menu icon never overlaps with the player's weapon loadout icon" The benefit is that it doesn't need to use a separate language like Lean which has the problem of code drift (you have to essentially maintain two codebases, the actual code and the proofs and ensure they match) Because Rust has this crate called kani which lets you directly write proofs in Rust So I can just import the layout function and write proofs on it! It's saved me so much time because it makes sure all UI stuff look good on mobile, desktop, etc Then the next step is what do you do when the proofs don't hold? You need to make taste-bound decisions on how to draw the UI on different dimensions so it looks good So when a proof fails on a certain configuration of screen dimensions (e.g. mobile), the agent has tools to reproduce this failing configuration and take a screenshot and send it to me so I can make a decision on how to change the layout to look good on that screen

介绍比Jev快50倍，在你设备上跑的laya-mlx! 只在你的设备上占用最高1G内存 Laya是一个开源的类似于Jev的,基于文本输出概率的分类系统 我将其移植到MLX,并且做了一些性能优化! 视频中就是这个模型在我的本地M3Max上玩贪吃蛇 这个模型能够以每秒决策60次的速度玩贪吃蛇！ github.com/mizorewww/laya-ml…

98.2% of Qwen3.8 27B my ass. I got hyped and gave it a real agent job right away. Build me an FPS in three.js, 6 hours on a 3090. My most standard and default prompt that I always use. It spent the first 32K tokens on a plan without writing a single file, then shipped a black screen, 2 shaders that dont compile and a player who spawns dead. And wrote "verified" in the final report. Ok, too hard. I gave it the easiest thing I have, a voxel pagoda garden in one html file. Video attached. 3 hours for THIS. On the way it deleted its own file and spent an hour debugging a raycaster nobody asked for. Where the 98.2% is in all this I have no fucking idea. Same Qwen3.8 27B, same pagoda task, ISTA-DASLab GSQ-RCO IQ2_XS on a 12 GB 3080 Ti gave me day and night, real shadows and koi fish in the pond. 8.4 GB on disk, real 2.50 bpw, 131072 ctx with q4_0 KV, 47 tok/s at 128K. 2.5 bits beats 2.13 bits by a lot when the 2.13 is this shit. Post with the video and the exact command in the replies. In the replies, I'll attach what a proper Qwen3.8 27B created in my hands. Bottom line: if you have at least 12GB VRAM, use ISTA-DASLab GSQ-RCO of all THIS.