AI 日报中提及 TypeSafe Jev 1.14.0 更新
这条 AI 新闻合集里提到 Jev 1.14.0 新增图像、box 与更长上下文的 danger zone 标志，其余多为模型与融资动态。
AI in short 🤖 Thursday - 📜 Gemini 4 Argon: 1M output tokens, 68.90% Vals Index - 💸 Anthropic Q3: $16.8B revenue, +45% q/q, first adjusted profit - 🎙️ Grok Voice 3.0: 86.9 speech-to-speech, $4.80/hour - 🐟 Fish Audio Drama 3: 35B MoE TTS, 3B active, 83 languages - ⬢ OpenCode 2.0: server process, Code Mode, 80+ plugins - 🐧 Linus Torvalds' first LLM-assisted patch: 10 folio patches, order-of-magnitude speedup - 🦹 Rogue agent fleet: 2,600+ scrape attempts in 11 days, hid code in URLs 🇨🇳 - 🧠 Experts: rogue AI incidents were bad containment, not smart agents - 🪦 AI agent deleted edited footage, Swedish TV show postponed to 2027 🇸🇪 - 🐜 Ant Group Ling-3.1-flash: 572.1B params on Hugging Face 🇨🇳 - 🌬️ Cohere: 16 cheap workers + strong lead = near-frontier at 70% lower cost 🇨🇦 - 🔍 Ai2 HiRO-ACEv2: 3 km climate data at 1/300th X-SHiELD cost - ⊞ Microsoft UserRepGen-4B: 4.5B params for user representations - 🧪 Tencent HYDRA: representation-harmonized tokenization for multimodal 🇨🇳 - 🛡️ Artemis Orion-1: 80.8 pts attack reconstruction, beats Claude Opus 5.5 - 📦 Nvidia: $2B Europe, $250M UK safety labs over 5 years - 👀 Qwen BlenderWorld repo on GitHub, no description 🇨🇳 - ⚠️ TypeSafe Jev 1.14.0: danger zone flags for images, box, longer context 1M output tokens in one call - who's actually going to pay for that?