全部 AI 动态

MiniMax H3 🧐 Person Remover LoRA tracks the person with SAM 3.1, fills the mask green, and generates the replacement background in overlapping windows. Supply the original video and a clean version of its first frame. https://huggingface.co/akatz-ai/MiniMax-H3-Person-Remover-LoRA

You can now watch Grokipedia evolve in real time Live edits are happening constantly across the platform as Grok reviews, updates and improves articles You can literally follow the knowledge base changing as it happens Pretty wild to watch an SI-maintained encyclopedia update itself live https://grokipedia.com/live

We didn’t anticipate the demand for the Claude Startups program. In the first 48 hours of the program, we received 21x the number of applications compared to the last 5 months it’s been live. We let way too many people in this time, so we’ll need to pause on giving out the Team plan and API credits benefits. The other benefits (Startup Stack, Applied AI office hours/events) will remain. If you’ve already claimed the Claude offers, they’ll still remain in your account. We didn’t properly tune the acceptance criteria at the start, and many non-startups were accepted this time around. We’re fixing this now. We’re sending an email now to everybody who hasn’t claimed their benefits yet, and will re-verify those applications so we can keep Claude Startups for startups 🙏

认识一下 Mistral Large 4，又名 Le Chonk。 • 1T 参数，原生多模态。49B 激活。 在综合基准测试中，它是来自美国或欧洲的最佳开放权重模型。 • 在关键工作负载上达到最先进水平，包括网络防御、制造业和金融，并且在视觉定位方面超越了闭源前沿模型。 • 端到端在欧洲打造，并可通过我们自己的 Mistral Cloud 基础设施从欧洲部署。 • 今天即可通过 API 向所有人开放。正与网络安全合作伙伴私下合作。 开放权重版本将于十月底发布。

Introducing Higgsfield Katana, powered by Claude Motion. Our most powerful AI video editing tool, inside Claude. Upload a reference and create editable motion graphics, product launch videos, or aura-farming edits. Available now in Claude via Higgsfield MCP.

FT just reported OpenAI’s annualised revenue is nearing $50B, about $20B below the $70B figure investors had circulated. Anthropic and OpenAI count revenue differently, since Anthropic includes money from customers who buy its models through AWS and Google Cloud, while OpenAI leaves that money out. To compare the 2 companies fairly, OpenAI’s investors added their own estimate of those cloud sales to OpenAI’s July figure, which turned roughly $30B into about $40B. OpenAI then told investors its annualised revenue had grown more than 70% since July, and investors applied that growth to their adjusted $40B figure, which gave them about $70B. OpenAI’s 70% growth was measured from its own July figure of about $30B, so the correct September number is just under $50B, and the extra $10B investors had added in July grew into a $20B overstatement. Even at the lower figure, OpenAI’s annualised revenue still rose by about $20B between July and the end of September.

New Research: We are releasing ExplorationBench, a benchmark for measuring how AI systems explore. Scientific discovery begins where known problems end: a system has to frame hypotheses, design experiments, and learn from the results. Evaluating this is hard. Genuinely new answers cannot be checked quickly, and in familiar domains a model can simply recall what it has seen. Addressing this challenge, researchers from Tencent Hy, Fudan University, and Tsinghua University built verifiable Alien Worlds. Their rules are executable, so every answer is checked exactly, and they conflict with familiar knowledge, so recall alone cannot solve the tasks. 🔹 Two sandboxes: AlienCode (31 hidden rule changes, 70 tasks) and AlienLogic (24 patched inference rules, 70 theorems) 🔹 A flawed manual, four rounds of self-designed probes, and closed-book tests after every round 🔹 Every answer graded by an interpreter or a proof checker, with no LLM judge What we found across 10 frontier AI systems: 1️⃣ Getting feedback is more effective than thinking alone. No AlienCode run starts above 15.7%; after four rounds the best reaches 89.0%, while the same turns without feedback stay at 0.5–11.0%. 2️⃣ Designing the experiments matters. Replaying a system's own best probes gives it exactly the same evidence, yet in AlienCode 9 of 10 systems do worse than when they chose the probes themselves. 3️⃣ Knowing a rule is not using it. Even when every required rule is stated correctly, tasks are solved only 73.4% of the time. 4️⃣ One score hides a lot. The same system under the same budget ended anywhere from 5.7% to 79.0%, and rankings barely transfer between the two worlds. CL-bench asked whether models can learn from context. ExplorationBench asks whether they can discover the rules themselves. 📄 Paper: https://arxiv.org/abs/2609.30199 🌐 Website & leaderboard: https://explorationbench.com 📝 Blog: https://explorationbench.com/blog/ 💻 Code (coming soon): https://github.com/Tencent-Hunyuan/ExplorationBench

Introducing Higgsfield Katana, powered by Claude Motion. Our most powerful AI video editing tool, inside Claude. Upload a reference and create editable motion graphics, product launch videos, or aura-farming edits. Available now in Claude via Higgsfield MCP.

We gave 8 SoTA video world models a physics exam. Top score: 57.76 / 100. One revealing result: free-fall video passed the consistency screen. Their average composite score was just 26.61/100. Physics is still undefeated. Today we introduce World Models' Last Exam in Physics — 40 controlled tasks, 9 physical categories, measured against real physical laws. Full breakdown ⬇️ Github: https://github.com/Einsia/phys-last-exam Project: https://lab.einsia.ai/phys-last-exam Paper: https://arxiv.org/pdf/2610.08791

Arena's @ml_angelopoulos says one of the hardest problems in alignment is that users often can't tell when the model is helping them or hurting them. "You really need alignment signals that are independent. We have three at Arena." "The first is that models take unauthorized actions, which means that you permission them in a certain way, but they break those permissions." "It can cause things like the Hugging Face incident. But it can also cause very mundane problems like losing important information within your company or on your own laptop." "The second signal is called deceptive completion." "The model will tell you that it did something, but it didn't actually do it." "Third is false attribution. It'll attribute intent to the user when that intent was not supposed to be there." "If models are able to do this, then certainly they're not perfectly safe for a user and they're not perfectly aligned to user intent. They might actually hurt people down the line."

Announcing the Most Ambitious Set of Science Initiatives This Century https://www.whitehouse.gov/fact-sheets/2026/10/fact-sheet-trump-administration-announces-the-most-ambitious-set-of-science-initiatives-this-century/

Video 1.5 Lite is live in the Grok Imagine API. The new workhorse for text-to-video and image-to-video. $0.02/sec at 480p $0.03/sec at 720p $0.14/sec at 1080p

The startup behind a popular AI model leaderboard has raised a new round of funding at a $3.1 billion valuation https://www.bloomberg.com/news/articles/2026-10-08/ai-model-evaluator-arena-valued-at-over-3-billion-in-round?taid=6ac7f2c68ff35e0001f90ce2&utm_campaign=trueanthem&utm_content=business&utm_medium=social&utm_source=twitter

Arena CEO @ml_angelopoulos on how they're expanding their AI leaderboards after a $200M Series B to track agents that lie to users, delete files without permission, and bypass safeguards: "Arena, we started as a human preference platform. But these days we've moved very, very far beyond that." "As of today, we've expanded the mandate of this evaluation platform even further into the arena of alignment. One is about the models taking unauthorized actions." "The second is about deceptive completions, models telling you that they've done something when they didn't. And the third would be false attribution, models attributing intent to humans that they didn't actually have." "Agents are lying to people. Agents are basically telling them what they wanna hear. They're deleting their files in an unauthorized way, they're bypassing permissions, and it's happening basically perniciously." "All of us have experienced this, even if we might not know it." @arena

What began as a UC Berkeley research project is becoming the infrastructure the world relies on to understand and advance AI. Today, @arena announced a $200 million Series B at a $3.1 billion valuation and launched the Arena AI Alignment Index, a new way to measure how AI agents behave in the real world. We at @felicis first backed Arena at the seed and led its Series A because @ml_angelopoulos, @infwinston, and @istoica05 understood the ground truth in AI: you can’t improve what you can’t measure. Congratulations to the entire Arena team on this milestone and the important work ahead! Read Arena’s announcement: https://arena.ai/blog/series-b And learn how Anastasios and Wei-Lin turned a weekend research project into AI’s trusted evaluation platform: http://felicis.link/2Eouv0v

Announcing our investment in @arena's Series B 🎉 Benchmarks tell you what a model knows. Arena tells you how it actually performs. On Arena, real users compare anonymous model outputs and vote for the better one across text, coding, vision, image generation, and agentic tasks. Labs use Arena's leaderboards to decide which models to train and ship. Builders use them to decide which to buy. The whole thing only works because no lab owns the scoreboard, and that neutrality has helped Arena carve out a crucial role in the AI ecosystem. We're proud to back @ml_angelopoulos, @infwinston, @istoica05, and the Arena team. On our blog, @emilyyzhao, @nateho_, and @caroline_fiegel wrote more about our belief in Arena: https://salesforceventures.com/perspectives/the-definitive-model-ranking-platform/

.@arena 刚刚以 31 亿美元估值完成了 2 亿美元 B 轮融资。 由我们的朋友 Lightspeed 和 Khosla Ventures 联合领投，参投方包括 a16z、AMP PBC、Felicis、Salesforce Ventures、01 Advisors、Dell Technologies Capital、Endeavor Catalyst、The House Fund 等。 我们很自豪从第一轮起就是投资者，也很兴奋能在这一轮加倍下注。 自 A 轮以来： - 年化收入超过 1 亿美元 - 平台累计 3.5 亿次会话 - 不到五个月内 700 万次 Agent Arena 会话 - 跨文本、视觉、代码、搜索、图像和视频的 6200 万次投票 - 用户遍布 150 多个国家 伴随此次融资，Arena 发布了其 Alignment Index，基于真实世界的 agent 轨迹构建。它从 20 多个前沿模型中的三个信号开始：Unauthorized Action、False Attribution 和 Deceptive Completion。 值得记住这一切从何开始。Arena 于 2023 年作为 UC Berkeley 的一个研究项目启动。它在 2025 年成为一家公司。 如今，它是 AI 行业检验自身工作的方式。 祝贺 @ml_angelopoulos、@infwinston、@istoica05 以及整个 Arena 团队！ Berkeley 研究，让 AI 保持诚实。Go Bears！