archerhume.comJev’s Architecture Unmasked — archerhume@kotsakis · sep 22Save1 saveBy probing Jev's API with 10k calls, the author infers Jev is a causal transformer (possibly sparse MoE) that reads shared state and outputs decision probabilities directly without text generation.More onJev55typesafe11#developer-tools4938#ai2911#software-engineering2268#machine-learning271RelatedIntroducing Clairic, a new System One class of extraction modelsCloudflare introduces Clairic, a fast, cheap System One extraction model for extracting, computing, estimating, and forecasting values from unstructured data.also on typesafe, Jev, #machine-learning, #ai, #developer-toolsAnalyzing Jev, a new AI model — Senko RašićJev, a new System One classification model derived from BERT, offers LLM‑like usability with far higher speed, lower cost, and type‑safe outputs without hallucination or fine‑tuning.also on typesafe, Jev, #machine-learning, #aiHow Jev works: calibrated decision modelsJev is a low-latency decision model that scores options via LLM forward passes to output calibrated probabilities without text generation.also on Jev, #machine-learning, #software-engineering, #ai, #developer-toolsGemini Decisions API: Fast multimodal decisionsGoogle DeepMind launched the Gemini Decisions API with a new model for fast, multimodal yes/no, choice, and rubric-scoring decisions in 150-250 ms at $0.10 per 1M input tokens.also on Jev, #machine-learning, #ai, #developer-tools