2 months agoAI News CN (Telegram)DeepSeek低调宣称小更新 结果用2折价格达到R1水平 在大模型竞技场测试KCORES中，DeepSeek-V3-0324代码能力得分328.3分，超越普通版 Claude 3.7 Sonnet（ 322.3 ...
2 months agoAI News CN (Telegram)🖼 [图片内容无法直接翻译，但根据图片内容推测，这应该是一张新闻报道的截图，其中可能包含文字和图表。以下是对可能存在的新闻内容的推测性翻译，基于常见的...
2 months agoAI News CN (Telegram)#Tip #Gemini Gemini 2.5 Pro 会默认进入思考模式，且无法通过简单的提示词关闭思考模式。 Gemini 2.5 Pro will ‘think’ by default, which cannot be turned ...
2 months agoAI News CN (Telegram)#Update #Gemini Gemini 网页版正在灰度推送 Gemini 2.5 Pro。 Gemini 2.5 Pro is being rolled out on the web. via AI Copilot - Telegram Channel
2 months agoAI News CN (Telegram)#Update #Grok Grok iOS App v1.0.57 已经支持编辑图片功能。 Grok iOS App v1.0.57 now supports Image Editing with Grok. via AI Copilot - Telegram Channel
2 months agoAI News CN (Telegram)Elevated errors for Claude.ai, Console, and the Anthropic API Mar 25, 05:07 PDT Investigating - We are experience increased error rates for some custo...
2 months agoAI News CN (Telegram)Deepseek 官方推荐温度设置 temperature 参数默认为 1.0。 我们建议您根据如下表格，按使用场景设置 temperature。 场景 温度 代码生成/数学解题 0.0 数据抽取/...
2 months agoHugging Face Daily PapersLEGO-Puzzles: How Good Are MLLMs at Multi-Step Spatial Reasoning?
2 months agoHugging Face Daily PapersFullDiT: Multi-Task Video Generative Foundation Model with Full Attention
2 months agoHugging Face Daily PapersSelf-Supervised Learning of Motion Concepts by Optimizing Counterfactuals
2 months agoHugging Face Daily PapersTracktention: Leveraging Point Tracking to Attend Videos Faster and Better
2 months agoHugging Face Daily PapersTokenHSI: Unified Synthesis of Physical Human-Scene Interactions through Task Tokenization