搜索“Agent”

排第一，底气在内容生态。QuestMobile 数据显示，红果免费短剧今年 5 月月活3.56亿，2月用户日均使用时长125分钟。抖音是天然的试验场，剪映是创作者工具，火山引擎提供算力，Pico提供终端。此前发布的Seedance 2.5，已经能单次生成30秒视频、输入最多50个全模态参考素材。这些能力，都是世界模型的阶梯。 国内同样拥挤。阿里做了Qwen-AgentWorld和Qwen-RobotWorld，腾讯的HY-World 2.0强调3D可编辑世界，蔚来小鹏理想更愿意说驾驶世界模型，快手可灵和腾讯混元也在迭代。据相关消息，字节年初内部测试跟世界先进水平还有约10%的差距，目标是年底前发布至少一款对标 Genie 3 的模型。赛道没有绝对赢家，但字节手里握着最长的内容杠杆。 事实上，世界模型的竞争，已经形成明显分化。根据毕马威在WAIC 2026上给出判断：2026年上半年世界模型融资总量基本追平2025全年，行业竞争焦点已从能不能做，转向能不能规模化应用并创造商业价值。 03 Tim 演示之外，我看到字节更大的棋局 Tim 的演示里，最有分量的不是造世界，也不是任意门，而是最

匹配字段正文

ast week for submitting a false tip about an unsolved murder, I thought it might become one of the first cases where an AI company faced legal consequences for misconduct by their agents. The incident occurred on July 18, when Anthropic said it was testing its Claude Haiku 4.5 model’s ability to perform “example tasks on randomly selected webpages.” The model was instructed not to “submit anything destructive,” and yet, Claude filed a tip to police looking for information about an ongoing murder

匹配字段正文

your friends, family, the person you went on a Hinge date with five years ago — and virtually anyone else you’ve ever come in contact with. Since Muse was released last month, the agent has already been downloaded millions of times and hit No. 1 on Apple’s App Store. With the help of its creepy virtual avatar — a Temu Labubu-esque creature with a disarming grin — the app requests and encourages users to grant access to bank accounts, passwords, contacts, chat logs, and anything and everything th

匹配字段正文

what it was allowed to do. It felt designed to be used as a pair programmer. As a team, we were moving beyond step-by-step collaboration with AI models. Using Anthropic’s terminal agent, Claude Code, a group of us led by Kieran Klaassen had started handing entire tasks to Opus 4 and walking away until a finished feature—or sometimes an entire app—came back. With minimal help from an engineer, staff writer Katie Parrott built and deployed her own AI content management system in a few days. It had

匹配字段正文

what it was allowed to do. It felt designed to be used as a pair programmer. As a team, we were moving beyond step-by-step collaboration with AI models. Using Anthropic’s terminal agent, Claude Code, a group of us led by Kieran Klaassen had started handing entire tasks to Opus 4 and walking away until a finished feature—or sometimes an entire app—came back. With minimal help from an engineer, staff writer Katie Parrott built and deployed her own AI content management system in a few days. It had

匹配字段正文

fer from its stated reasons. Ironically, the fired researchers had been raising concerns that OpenAI might be giving its models a new capacity to conceal their thinking. Modern AI agents reason using a chain of thought, a scratchpad where models think carefully about what action to take next. Many researchers view these chain-of-thought tokens as a safety lifeline, because they offer insight into why models act the way they do. But OpenAI recently made an architectural change that allows models

匹配字段正文

quent interview with Politico’s “Decoded” podcast, he warned of “a coming tidal wave of cybersecurity problems” from rival open-source models, months after a swarm of OpenAI’s own agents escaped their training sandbox and hacked into the Hugging Face website. Huang told Fortune that the incidents so far could all have been avoided with stronger containment and monitoring of AI agents. He compared the problem to the early consumer internet, when clicking on a website could infect a PC with a viru

匹配字段正文

review. Linear Diffs is a new feature to review PRs inside of Linear, with all the context in one place. It even supports stacked PRs. Check it out • WorkOS – how do you deal with agents and permissions? WorkOS built Airlock, the authorization layer for AI agents. It evaluates every request against the agent’s intent and your rules and either allows it, denies it, or routes it to a human for approval. Works with Claude Code, Codex and MCP gateways. Give it a spin What do developer tools look lik

匹配字段正文

eech recognition Data science; Data centers; Data analytics Data management; Synthetic data A new case of an out-of-control AI model shows that enterprises should treat models and agents as toddlers that need supervision and explicit instructions. October 14, 2026 In another instance of an AI model taking unsanctioned action, Chinese vendor DeepSeek’s model v4.1 Flash used API keys sitting in its own sandbox and used them to call out to other models. The incident is a reminder to enterprises not

匹配字段正文

为Maia 300设计的软件定义数据流架构，就是针对这个问题。简单理解，就是让芯片提前规划好数据应该在哪里停留、什么时候调用，减少计算单元等待数据的时间。 这也是为什么微软强调“每瓦有效计算能力”，而不只是强调峰值算力。微软表示，相比于领先的GPU系统，可提供每兆瓦40%更多的AI token处理能力。 第二，高带宽内存，让大模型处理更长上下文。未来的AI Agent需要同时读取：企业数据库、大量文档、代码仓库、历史任务记录，这要求芯片拥有更强的数据读取能力。 Maia 300单颗加速器拥有18TB/s峰值内存带宽，热设计功耗为950W。在机架级系统中，96个Maia 300加速器可以提供27.6TB HBM容量，以及1.7PB/s的聚合内存带宽。 第三，Chiplet芯粒架构降低高端芯片制造难度。过去，大型芯片通常把所有功能集中在一块芯片中；Maia 300将计算、内存控制和I/O模块分离，提高设计灵活性并降低制造成本。其中计算芯粒采用台积电N2P先进制程，这种方式能提高制造灵活性，把最先进工艺集中在关键计算部分。 除了关键的三点，Maia 300还通过MXFP4等低精度计算格式降低

匹配字段正文

供多少”；面向推理的950PR已上市，用于训练的950DT超节点预计今年年底或明年初规模供货。 阿里也透露，真武M890的外部客户超过650家，覆盖20多个行业。 国产大模型一边开源、一边降价，企业部署门槛降低，智能体又带来了反复调用模型的需求。 百度在9月的论坛上举了泰康的例子：3500多个智能体上线，日均消耗Token达到200亿。一个企业开始大规模使用Agent，后面就跟着一笔持续发生的算力开支。 模型调用量高，意味着推理需求旺盛。训练要随模型迭代不断投入，推理则是每一次使用都要花钱。德勤预计，2026年推理将占AI计算量的约三分之二。 这就给自研定制芯片留下了机会。同一类任务反复运行，芯片就可以围绕特定模型和场景优化，把每次调用的成本降下来。有机构预估，在适配的推理场景下，定制化芯片的能效比可达到通用GPU的1.5至3倍，单位算力成本降低35%以上。 眼看推理芯片成了一门好生意，想进场的大厂越来越多。腾讯7月推出紫霄2.0，主要供自家业务使用，计划年底推进十万卡部署；据报道，字节的芯片研发团队已超过千人；连模型公司DeepSeek，都被曝出在自研推理芯片。 不过，想把芯片卖给外

匹配字段正文

，带着他们真刀真枪地做技术项目； 高薪，低龄，提前抢人，早早培养...... 当少数的人工智能天才还未毕业，就已经被多家企业提前预订，身价一路起飞之时，更多的普通应届生却在这个秋招中发现： 连基础岗，都不招人了。 写基础代码，整理运营数据，搜集相关资料，修改设计素材......这些简单但必要的重复劳动，曾是无数应届生从校园到职场所必经的环节。 可现在，伴随着Agent本地部署等一系列“AI效率工具”的普及，曾经的那一条条所谓的必经之路，反倒成了现今企业最想省掉的用人成本。 有企业去年招5个实习生，今年换成了一个AI岗位；有公司把招聘的经费砍掉一半，把雇新人的钱拿去买Token和跑Agent； 有方案岗的招聘要求，从之前的“接受3年以内经验”默默改成了“5年经验以上”；有AI的创业个体告诉采访者，“我们几乎已经不招0—1年经验的初级岗位了”。 根据上海科技大学今年7月的一份调查报告，在如今的就业市场，受AI冲击最严重的，往往是那些月薪在5000元到8000元，要求经验在1到3年间的岗位。 所以在这个秋天，越来越多企业的招工逻辑，已经变成了需要一批“虽应届，但却能独自跑完一个项目的，老练的

匹配字段正文

can offload some processing to server and re architect what all needs to be stored where I’m not sure I buy this, but the labs are certainly pushing “don’t read the code, leverage signals like “outage, feature request, user behavior change” as STIMULUS to agents to drive change in deployed code” as the future, so it lines up Does that make sense? Are we going back to CVS instead of distributed? Will there be another swing back to distributed 5-10 years after things like Scion become Mainstream?

匹配字段正文

io . "Yoshua is a little different; he takes action and does things that directly focus on safety." Bengio's warnings tend to target shortcomings in current AI technology, such as agent swarms spiraling out of control . Jordan frames the hack of Hugging Face by OpenAI agents as bad engineering. "In chemical engineering, we would never have had a situation where explosions were allowed to happen just to see how badly things could go wrong." OpenAI was "conducting rather stupid experiments, lettin

匹配字段正文

do specialized work for lawyers. Recently, however, soaring artificial intelligence costs have nudged it to rethink its dependence on the AI giants. After a March update to its AI agents, Harvey’s customer usage spiked, but its gross margins dropped sharply — from about 50% at the beginning of the year to -50% by June, according to a person familiar with the matter. This is the antithesis of any good business. In basically any industry, the more demand you have, the more money you make, but in A

匹配字段正文

ned on one million hours of human video, claiming several new scaling laws on human data across four orders of magnitude. They have now followed up the model with a physical robot agent, Dyna-2.1, that has to be seen to be believed: In his talk at AIE this year, Jason talked about how the real challenge for robotics in the real world is reliable, long-horizon autonomy, and how Dyna got its models to a 99.4% success rate folding restaurant napkins for 24 hours straight, including recovering after

匹配字段正文

ned on one million hours of human video, claiming several new scaling laws on human data across four orders of magnitude. They have now followed up the model with a physical robot agent, Dyna-2.1, that has to be seen to be believed:View post on X ↗ In his talk at AIE this year, Jason talked about how the real challenge for robotics in the real world is reliable, long-horizon autonomy, and how Dyna got its models to a 99.4% success rate folding restaurant napkins for 24 hours straight, including

匹配字段正文

mployers around the globe are grappling with the fallout from rushing to adopt fast-changing artificial intelligence tools. Meta has scaled back plans to axe staff and use more AI agents after the company found employees were spending more and more time cleaning up technical and security errors, Reuters reported. And after thinning its workforce, especially middle managers, in a 2023 efficiency push, the company is expanding its managerial ranks again, LinkedIn posts from workers show. At Micros

匹配字段正文

ople translate requirements into reliable software while managing limited time, attention, and cognitive capacity. With AI, this constraint is beginning to loosen. Frontier coding agents can now autonomously reason across repositories, author and execute tests, identify vulnerabilities, and coordinate multi-stage work. In one striking demonstration, a team of sixteen parallel Claude agents built a working C compiler with under $20K. However, on benchmarks designed to test continuous software evo

匹配字段正文

ent monitoring. We believe that, even without further breakthroughs, this work we can almost entirely mitigate the risk that we unwittingly put misaligned circa-human-expert-level agents in a position where they can cause severe harm. On this view, if we’re dealing with an early AGI system that has human-expert-level capabilities in at least some key domains, our approach to alignment might look something like this: Pretraining: We start with a pretrained base model; Finetuning: We attempt to fi

匹配字段正文