Introducing Contrastive Language Model (CLM): an ultra-fast System One Model trained with a contrastive learning objective that connects states and actions. CLM-8B is pre-trained on internet-scale data and delivers up to 9× faster inference than Jev ⚡ while achieving comparable performance across computer-use, gaming, and tool-calling tasks. With lightweight fine-tuning, CLM-8B sets a new SOTA on challenging agentic coding benchmarks, such as DeepSWE (81.6%) and Terminal-Bench 2.1 (87.6%). In contrast, Jev fails to serve as an effective verifier for these long-horizon tasks. We also build an efficient training and serving infra for CLMs by disaggregating states and actions, allowing their embeddings to be cached and reused independently. This substantially reduces inference latency in settings where the state evolves continuously while the action set remains fixed. Finally, we establish scaling laws for CLMs and show that the test contrastive loss decreases predictably as a power law in training compute, model size, and dataset size. 📄 Blog: contrastive-lm.notion.site 💻 Code: github.com/Contrastive-LM/CL… 🗣️ Discord: discord.gg/5dAQEDJBs 🤗 Data & Models: huggingface.co/Contrastive-L… More details on CLM’s architecture, data recipe, and scaling laws in the thread below 🧵

It's fucking insane how bad LLMs are at producing good code. It's like, they are actively trying to add complexity to your code base. They are so, so bad at it

to believe 7 is prime, first you need to buy that it's not divisible by 2. then another patch to rescue your theory: it's also not divisible by 3. then on top of that, somehow not by 4 either. and wow guess what, not by 5 either. soon it's a whole precarious tower of speculation

My wife grew up on the metric system, and she says the mile is the most ridiculous unit ever invented. For years, she had me. But history claps back.... The word comes from Latin, mille passus, a thousand paces. A Roman pace was two steps, counted each time the same foot hit the ground. A thousand of those made a mile. Along the Appian Way, built in 312 BC, stone markers stood at every thousand paces, counted outward from a gate in Rome's old wall. A traveler read the distance off the road itself. The strange number came much later. In 1593, Parliament under Elizabeth I set the mile at 5,280 feet so it would divide evenly into furlongs, the length of an English farmer's furrow. So the odd number is a Roman soldier's stride, adjusted to fit an English field. I love that. Every time you say "mile," you are still counting the steps of a legionary marching south out of Rome..... ...on the Appian Way that you can still walk today, counting the old mile markers as you go The poet Statius called that road the queen of long roads.

JEV actually doesn't even run deterministically Exact same prompts give you different probabilities when you run it multiple times The ORDER of the choices DRASTICALLY changes the output probs I am more and more confused by the their "no-hallucination" claim

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…

Mathematician Terence Tao: "we have to slow down AI. the pace is insane, and there's no reason to be this fast — no reason at all" It's amazing how willing we are to change everything without any idea what happens afterward These are extremely nonlinear dynamics

you can make any open source model behave like jev with just a bit of inference engineering. it's shockingly easy. to prove it, we built a new endpoint we're calling deepseek-v4.1-flash-jev. see the demo below. here's how it's done: sglang (an inference engine) offers a scoring endpoint in addition to the normal generation one. in scoring mode, given an input & set of possible answers, it forces the model to produce probabilities for each one. example: > input: what is most common letter in abcccde? > possible answers: a, b, c > output: (c, 0.9), (b, 0.0.5), (a, 0.05) getting the above behavior instead of streamed output is as simple as using sglang's /v1/score endpoint instead of /generate. there's just one other trick required. for deepseek, you have to add a closing think tag before the response. this forces a direct answer instead of a reasoning trace. if you want reasoning, you can do that too, but imo that makes things too slow to be worth it. dsv4.1 flash is not as good as jev, but if we had enough spare compute to experiment with this same approach for a larger model then i think the decision quality would be at least as good, if not better. also, somewhat unrelated, i think decision-making models kill all prospecting & sourcing work. i would have absolutely killed to have jev or similar when i was recruiting @mintlify. absolutely incredible.

Yesterday I said Jev would open a ton of doors... 24 hours later, this exists. Cua built a 2.8MB model that scored 99.7% on their form-filling eval. Hosted Jev scored 83.6%. Not to mention it's FREE and only 706K parameters. Small enough to run locally with not even 1gb or ram. Fast enough to make decisions in one pass. And specialized enough that your agent doesn't need to call a giant LLM for every tiny action. Think about what this unlocks. Every repetitive computer task could eventually get its own tiny specialist: • forms • CRM updates • data entry • browser actions • document routing • UI decisions Then one powerful agent just routes work between them. We are going to see some ridiculous stuff built from this as well.

The fly figured out ways to solve and compress chess that are far beyond what we could possibly comprehend. I had to stop training it because it was destroying Opus 5. Demo and model available on @huggingface