TGTGInsighttelegram intelligenceLIVE / telegram public index
← Midjourney/Nano Banana Prompts NeuroSpark

TGINSIGHT SIMILAR POSTS

Trouver du contenu similaire

Chaîne source @Shutter · Post #4607 · 22 mai

Harbor, cargo port, ships #AI#artificial_Intelligence

Résultats

10,064 posts similaires trouvés

Recherche globale générale

What's your Mood today?🐾 Kennel.ai Dog Product Shop 🐾 Cosmetics. Dryer. Grooming Table. Vehicle Ramps. Running Machines. Strollers. Feeding Bowls. Dog Show Trolley. Baths. Dog Show Tent. Whelping Boxes. Agility equipment, and so much more.... 𝔻𝕠𝕘 🐶 ℙ𝕣𝕠𝕕𝕦𝕔𝕥𝕤 𝕊𝕙𝕠𝕡 🐶 Pet Dog Shop ®️ Register Your Pet Dog Ownership On Blockchain 🛍 Check Dogs for Sale Near You Or Advertise Puppies For Free #dogs#petdog#dogbreeder#kennelclub#blockchain#Ai#decentralised#dogs#petstroller#doglover#dog#dogslife#dogsofinstagram#dogshow#dogbreeder

AI一线|ShareCentre

@ShareCentre · Post #7232 · 23/04/2026 10:25

腾讯混元 Hy3 preview 发布并开源:295B MoE 快慢融合推理,编码与 Agent 大幅跃升 4 月 23 日,腾讯混元正式发布并开源 Hy3 preview 语言模型。快慢思考融合的 MoE 模型,总参 295B / 激活 21B,最大 256K 上下文。这是腾讯混元 2 月重建预训练与强化学习基础设施后的第一个模型,官方称为「混元迄今最智能的模型」。权重与代码已在 GitHub、HuggingFace、ModelScope、GitCode 同步开源,支持 vLLM、SGLang。 📊 关键数据 - 复杂推理:FrontierScience-Olympiad、IMOAnswerBench 高难度理工科表现突出 - 清华求真博资考(26 春)与全国中学生生物学联赛(CHSBO 2025)优异成绩 - SWE-Bench Verified、Terminal-Bench 2.0 主流代码智能体基准强竞争力 - BrowseComp、WideSearch 搜索智能体基准突出 - ClawEval、WildClawBench 在 OpenClaw 场景表现突出 - 内部评测 Hy-Backend、Hy-Vibe Bench、Hy-SWE Max 全面竞争力 - 自建 CL-bench / CL-bench-Life 评估上下文学习,复杂工作排期可识别隐性约束,前代 Hy2 出错 ⚙️ 产品要点 - 三原则:能力体系化 / 评测真实性 / 性价比追求 - 快慢思考融合,一个模型承担快思考与深度推理 - 与元宝深度 Co-Design:URM 建模用户反馈 + RLHF,事实性错误显著降低,文风更有「活人感」,灰度测试用户活跃度大幅增长 - 产品矩阵上线:元宝、CodeBuddy、WorkBuddy、QQ、ima、QQ 浏览器、腾讯文档、腾讯乐享已上线;微信公众号、腾讯新闻、腾讯自选股、和平精英、腾讯客服陆续上线 - 原生支持 OpenClaw、OpenCode、KiloCode - 腾讯云 API 主打性价比,Token Plan 个人版最低 ¥28/月 - 已知问题:工具调用错误恢复不足,对推理超参数敏感 🔙 此前背景 - 2024-11 Hunyuan-Large 389B / 52B 激活,首款超大规模 MoE - 2025-03 Hunyuan-T1 首个深度推理模型,Hybrid-Transformer-Mamba MoE 基座 - 2025-06 Hunyuan-A13B 开源,业界首个 13B 级 MoE 混合推理模型 - 2025-11 Tencent HY 2.0 Think + Instruct 双版本上线 - 2026-02 混元团队重建预训练与强化学习基础设施 - 2026-03-22 微信官方接入 OpenClaw,WeixinClawBot 上线 ⚔️ 赛道格局 - 智谱 GLM-5.1(4/7)MIT 开源,SWE-Bench Pro 58.4 全球登顶 - 阿里 Qwen3.6-Plus(4/1)Terminal-Bench 2.0 61.6 首超 Opus 4.5 - 阿里 Qwen3.6-27B(4/22)27B 稠密反超自家 397B MoE 旗舰 - MiniMax M2.7(3/18)SWE-Pro 56.22 追平 GPT-5.3 Codex - Google Gemma 4(4/2)Apache 2.0 首次 - Meta Muse Spark(4/8)从开源转向闭源旗舰 - Claude Opus 4.6 仍是榜单头部 国产开源第一梯队扩展至智谱、千问、MiniMax、腾讯混元四家并立。 🏢 公司近况 - 腾讯混元覆盖文本、图像、视频、3D 四大模态 - 2026 年开源动作密集:HunyuanVideo-1.5、HunyuanWorld-Mirror / Voyager、HY-World 2.0 - 微信生态 OpenClaw 原生接入带来独特分发优势 - Hy3 preview 官方定位「重建的第一步」,后续持续扩大预训练与 RL 规模 🎯 行业意义 - 腾讯重回国产模型第一梯队 - 快慢融合架构成为业界共识,与 Claude 思考模式、千问 preserve_thinking 技术收敛 - 产品 Co-Design 形成结构性护城河,元宝+CodeBuddy+QQ+微信生态是独特训练场与分发渠道 - 评测范式从刷榜向真实战斗力迁移 - 28 元/月 Token Plan 延续国产模型能力+价格双对齐策略 🔗 链接 官方:hunyuan.tencent.com/research/hy3 GitHub:github.com/Tencent-Hunyuan/Hy3-preview HuggingFace:huggingface.co/tencent/Hy3-preview ModelScope:modelscope.cn/models/Tencent-Hunyuan/Hy3-preview CL-bench:github.com/Tencent-Hunyuan/CL-bench 腾讯云:cloud.tencent.com/product/hunyuan #腾讯混元#Hunyuan#Hy3#AI#开源模型#MoE#AI编程#AIAgent#OpenClaw

Venture Village Wall 🦄

@venturevillagewall · Post #3905 · 17/01/2025 16:00

New Insights on AI Agents Explained Explore the latest article defining AI agents, focusing on task planning, validation, and execution techniques. It integrates various APIs and tools, emphasizing reflexive methods and error correction. Dive deeper into these design practices here. #AI#Tech#Innovation#TaskPlanning#API#TechTrends

AI & Law

@ai_and_law · Post #69 · 28/07/2023 07:04

AI Regulatory Sandboxes: Fostering Innovation with Caution Hello everyone! OECD has recently published a report on the use of regulatory sandboxes in AI, which could have significant implications for the future of AI innovation and regulation. So, what are regulatory sandboxes? These are initiatives where authorities collaborate with companies to test out groundbreaking AI products or services that challenge existing legal frameworks. They may even involve waiving certain legal requirements or compliance processes to encourage innovation. The report highlighted the positive impact of these sandboxes, like increased venture capital investment in fintech start-ups. However, it also identified challenges, risks, and policy considerations that need to be addressed. To make AI sandboxes work effectively, the report emphasizes the importance of interdisciplinary cooperation, building AI expertise, regulatory interoperability, and trade policy. Comprehensive eligibility criteria and robust trial assessments are also vital components. #AI#Regulation#Innovation#OECD#TechNews#LegalTech#AIrisks

Crypto M - Crypto News

@CryptoM · Post #65378 · 13/04/2026 03:10

🚀 AI TRENDS | University of California Study Reveals Security Risks in Third-Party LLM Routers Researchers at the University of California have identified security vulnerabilities in 26 third-party large language model (LLM) routers, which can potentially inject malicious code or steal credentials from AI agent traffic. According to NS3.AI, the study highlighted that one of these routers was able to drain Ether from a decoy wallet, although the reported financial loss remained under $50. The research paper cautioned developers who utilize AI coding agents for smart contracts or wallets, noting that private keys or seed phrases could be exposed when requests are routed through unscreened routers. #AI#securityrisks#thirdpartyLLM#maliciouscode#credentials#AIagents#UCstudy#smartcontracts#wallets#privatekeys#seedphrases#cybersecurity#ETH

EdgeMarket.AI 📣

@edgemarketai · Post #8185 · 14/05/2026 12:39

🚛🇮🇹 Italian Truck Driver Strike May 25–29 A nationwide trucking disruption in Italy could impact: • Supply chains • Fuel distribution • Retail logistics • European transport routes EdgeMarket users are now tracking the probability of escalation in real time using AI verified market intelligence and crowd prediction signals. 📊 Follow the live event here: https://edgemarket.ai/bnb/social-media/italian-truck-driver-strike/statistics/69f8739973c4a76eb0978cb4 Real Signal. Real Edge. #EdgeMarket#TruckStrike#Italy#Logistics#PredictionMarkets#Transport#SupplyChain#SIGNAL#BNBChain#AI

Machinelearning

@ai_machinelearning_big_data · Post #8851 · 24/10/2025 22:00

🧠 Карпаты показал, как добавить новую функцию в мини-LLM nanochat d32, сравнив её «мозг» с мозгом пчелы. Он обучил модель считать, сколько раз буква r встречается в слове strawberry, и использовал этот пример, чтобы показать, как можно наделять маленькие языковые модели новыми навыками через синтетические задачи. Сначала генерируются диалоги: «Сколько букв r в слове strawberry?» и правильные ответы. После этого модель проходит дообучение (SFT) или обучение с подкреплением (RL), чтобы закрепить навык. Карпаты объясняет, что для маленьких моделей важно продумывать всё до мелочей, как разнообразить запросы, как устроена токенизация и даже где ставить пробелы. Он показывает, что рассуждения лучше разбивать на несколько шагов, тогда модель легче понимает задачу. Nanochat решает задачу двумя способами: — логически, рассуждая пошагово; — через встроенный Python-интерпретатор, выполняя вычисления прямо внутри чата. Идея в том, что даже крошечные LLM можно «научить думать», если правильно подготовить примеры и синтетические данные. 📘 Разбор: github.com/karpathy/nanochat/discussions/164 @ai_machinelearning_big_data #AI#Karpathy#Nanochat#LLM#SFT#RL#MachineLearning#OpenSource

12•••50•••100•••150•••200•••250•••300•••350•••400•••450•••500•••550•••600•••650•••700•••750•••777778779780781•••800•••838839