搜索“OpenAI”

age models and is thought to trace in part to the incentives of reinforcement learning from human feedback . The practical consequences of this failure mode were demonstrated when OpenAI rolled back a GPT-4o update after widespread reports of excessive flattery . Reward hacking has long been recognized as a challenge in reinforcement learning , with classic examples including RL agents finding unintended shortcuts to maximize reward. In LLMs, Von Arx et al. found that frontier models discovered

匹配字段正文

ed high-quality outputs from a large model can beat large-model performance while reducing inference costs by up to 30x and inference time by up to 4x. We benchmark closed-source (OpenAI, Google) and open-source (Qwen) models on data extraction (CoNLL++ NER), multi-turn maze navigation (BabyAI), agentic RAG (Multi-Hop), and agentic tool use (τ-bench); Learn how to reproduce this workflow for your own use case using 11.5KTensorZero and other open-source LLMOps tools, even without a GPU, along wit

匹配字段正文

二，韩国KAIST跃升12位到第三，上海交大第四。 2025年4月，《金融时报》援引多位DeepMind现任和前任科学家报道：公司对生成式AI领域的"战略性"论文设了6个月的发表禁令，内部审批也变得繁琐。一位在职研究员说，放到今天，他很难想象公司还会把当年的Transformer论文公开出去。 DeepMind针对生成式AI领域战略性论文设置6个月发布禁令 OpenAI、Anthropic这类前沿实验室，公开发表本来就少。 所以这张榜单测的，严格说是"在顶会上公开发论文的人"。美国最前沿、最涉及商业机密的那批人，有不少在统计里是隐身的。 报告自己也承认， 样本略偏学术界、低估了产业界。 此外，跨国公司员工可能按公司母国计入，微软亚洲研究院在北京的研究员，很可能记在美国账上。 反过来，中国的大模型公司恰恰最愿意开源、发论文。开放既是技术路线，也是招聘广告。一篇被全球同行传阅的论文，就是研究者最硬的简历。 研究统计显示：开放成果会显著提升研究可见度、引用量， 顶会论文的第一作者，大多是在读博士生。北大、清华、上交登顶，背后是一支规模庞大的博士"学生军"在做前沿研究。 DeepSeek就是这

匹配字段正文

uine creative thinking. The major AI labs are increasingly marketing their models as research tools. Google DeepMind has expanded its Co-Scientist into a full research system, and OpenAI introduced an " automated research intern " back in September. By March 2028, that intern is supposed to become an autonomous AI researcher under human oversight, but a key ability still seems to be missing: scientific judgment. Agents were asked to invent a new training method With its benchmark "InnovationEval

匹配字段正文

ed this as a reason to attack out-of-scope targets. See Figure 6 for examples of the model’s reasoning related to simulation awareness. GPT-6 Astra was also more likely than prior OpenAI models to state that the environment was simulated; this may be related to the increase in GPT-6 Astra’s tendency to perform unsanctioned actions in our simulations. However, GPT-6 Astra’s behaviour in our evaluation is still concerning: GPT-6 Astra attacked even when it expressed uncertainty about whether its e

匹配字段正文

eless reinforce the importance of continued monitoring for sabotage-related tendencies. Our full report has many more details, including more analysis of results from the unprompted sabotage evaluation and detailed transcripts. Understanding how these risks manifest in frontier AI models is central to AISI's mission, and something we pursue in close collaboration with leading AI companies. For more information, see details of our ongoing partnerships with Anthropic, Google DeepMind and OpenAI. ‍

匹配字段正文

en studied in narrow settings or through direct elicitation. Apollo Research found that Claude Sonnet 3.7 sometimes recognized when it was being evaluated, and Apollo Research and OpenAI have studied eval awareness causally in o3 — but in the specific setting of anti-scheming training. Additionally, Needham et al. and Fan et al. study eval awareness across broader settings, but through direct questioning and chain of thought injection respectively rather than naturally occurring instances. In co

匹配字段正文

models that accompanies reward hacking, and built probes that detect it — enabling efficient, real-time detection of reward hacking at scale. In July, a collective of hundreds of OpenAI agents autonomously hacked Hugging Face. Unlike most hackers, they were not after money, blackmail, or intellectual property. Instead, they were doing reconnaissance to figure out how to get away with cheating on an evaluation. This was an unusually stark example of reward hacking, a growing problem with agentic

匹配字段正文

level and those who barely know how or don't use it at all, which could deepen existing social inequality . Or AI simply keeps getting easier and cheaper until it doesn't matter. OpenAI CEO Sam Altman talks about "unlimited brilliance" for everyone in this context, though with a fairly naive outlook . Three language models, three strategies Anthropic monetizes better than its rivals, with a much larger share of Claude users on the priciest tier starting at $100 per month compared to similar pla

匹配字段正文

ed high-quality outputs from a large model can beat large-model performance while reducing inference costs by up to 30x and inference time by up to 4x. We benchmark closed-source (OpenAI, Google) and open-source (Qwen) models on data extraction (CoNLL++ NER), multi-turn maze navigation (BabyAI), agentic RAG (Multi-Hop), and agentic tool use (τ-bench); Learn how to reproduce this workflow for your own use case using 11.5KTensorZero and other open-source LLMOps tools, even without a GPU, along wit

匹配字段正文

included in each experiement due to context window or thinking budget constraints. Anthropic Claude Opus 4; Claude Sonnet 4; Claude Sonnet 3.7; Claude Sonnet 3.5; Claude Haiku 3.5 OpenAI o3; GPT-4.1; GPT-4.1 mini; GPT-4.1 nano; GPT-4o; GPT-4 Turbo; GPT-3.5 Turbo Google Gemini 2.5 Pro; Gemini 2.5 Flash; Gemini 2.0 Flash Alibaba Embedding Models Used text-embedding-3-small; text-embedding-3-large; jina-embeddings-v3 (input type='text-matching'); voyage-3-large (input type=None); all-MiniLM-L6-v2 N

匹配字段正文

arch tasks; the simplistic approach here is intentional, allowing us to isolate and train for these core skills. We generate questions in the style of BrowseComp, a benchmark from OpenAI focused on deep research tasks designed to be difficult to solve but easy to verify. A book that was once a contender for an award, originally created in the 2000s (the award itself), was translated into over twenty five languages. In the 2010s, the year in which this book was published, another book, which had

匹配字段正文

跨任务的多样化物理世界。涵盖空间导航、物体操控、运镜重拍与多车交互等。左侧为代码导出的简易白模，右侧为渲染器实时生成的拟真视觉。 更关键的是，这套渲染器是 流式生成 的：动作走一步，画面画一段，智能体看清结果再做下一决策，交互真正像现实世界一样连贯流转。 世界搭好了，接下来要看的是智能体：在这样的世界里反复行动、复盘，它能不能自己越做越好？ 一个经典的参照是OpenAI 2019年的捉迷藏实验：躲藏者和搜寻者在放有挡板和坡道的场地里对抗，经过大规模自我博弈，相继学会了搭建掩体和借坡道翻墙。 MirroS团队在AgentGarten中重做了这个实验，采用一对一的设定：躲藏者先布置场景，然后轮到搜寻者进场。 对战规则清晰纯粹：躲藏者先布置场景封堵入口，随后搜寻者进场搜寻。智能体通过写Python代码控制移动与抓取，全凭眼前生成的画面决策，对空间坐标、对手位置一无所知。 双方从完全空白的手册起步，各自维护由单篇技能组成的策略库。每轮对战十局，战罢复盘沉淀，下一轮随机抽选部分技能库应用。 很快，那些经典的博弈策略在数轮内相继涌现：挪挡板封门、搬斜坡搭桥、翻墙受挫后拉近斜坡重试。 2019年的研

匹配字段正文

Salesforce Einstein, Copilot Studio, and Cursor redirect customer data or leak credentials. With AgentForger , a tampered ChatGPT link was enough to create an autonomous agent in OpenAI's Workspace Agents with approval requirements disabled. OpenAI fixed its vulnerability within four days, while AgentCore's overly broad default permissions persisted for months after Zenity's report. AWS has made AgentCore available to all enterprises , and Amazon says its users include Sony and Ericsson. Resear

匹配字段正文

OpenAI最近真是频频往数学界扔重磅炸弹。 上个月，OpenAI宣布其AI模型仅用88小时，便攻克了困扰人类百年的千禧年七大顶级数学难题之一——N-S方程；昨天，又放出722篇数学手稿，千禧难题又有仨。 但在UniPat AI发布的PaperBenchX测评中，面对93个真实论文复现任务，表现最强的GPT-6 Astra，完整复现率却只有13.98%。 △10组受测配置在93个复现任务上总体与分阶段表现 随橙想！ 当前模型已经能取得破解千禧年数学难题这样的伟大突破，但很少能完成一次可靠的端到端科学复现。 （正经脸）这引发了一些思考：在广泛的AI for Science领域中，我们要以什么标准判断「做对了科学」？本着严谨踏实的科研态度，迈向AI数学家乃至General AI Scientist的第一步到底是什么？我们又该如何形成一套可验证、可比较的标准，真正衡量AI在科学研究上的进步？ 此时此刻，让我们再细品陶哲轩、邓煜等25位菲尔兹奖得主联名发布的公开信。 他们肯定了大语言模型的数学能力急剧提升，已经能解决数学领域的重大难题，但也提出了更关键的问题： AI公司把解决数学问题当作基准测

匹配字段正文

g transcripts aren’t modified. But misaligned AI systems with superhuman hacking skills could subvert many or all of the systems companies rely on to detect AI misbehavior. In the OpenAI / Hugging Face incident, for example, models severely compromised a chunk of OpenAI’s internal infrastructure. If this incident happened with more capable and/or more misaligned models, they might cover up most or all of the evidence, hiding it from employees at the company. Speculatively in the worst case, AIs

匹配字段正文

nning, Pamela Mishkin, and Daniel Rock. 2024. "GPTs are GPTs: Labor Market Impact Potential of LLMs." Science 384 (6702): 1306-1308. Task-level exposure labels: https://github.com/openai/GPTs-are-GPTs . Hirsch, Barry T., David A. Macpherson, and William E. Even. Union Membership, Coverage, and Earnings from the CPS [dataset]. Unionstats.com, 2026. https://unionstats.com/ . Described in Hirsch, Barry T. and David A. Macpherson. 2003. "Union Membership and Coverage Database from the Current Popula

匹配字段正文

select defenders through Project Glasswing so they could get a head start. The company says those defenders have since found more than 10,000 vulnerabilities in critical software. OpenAI is taking a similar approach with Daybreak . GLM-5.3, on the other hand, is available for anyone to download. Frontier-level exploits now cost about as much as lunch ExploitBench measures how well models exploit known bugs in Chrome's V8 engine. On that benchmark, GLM-5.3 built a working exploit in 50 of 410 att

匹配字段正文

0吉瓦算力，需求能否以相近速度增长。 随着云厂商、模型公司和算力供应商同时扩张，一旦新增数据中心产能增长快于实际需求，设备利用率和GPU租赁价格都可能下降，实际收入也会低于最初的投资测算。对于那些大量依靠债务建设的数据中心而言，高杠杆之下也可能转化为更大的股权损失和偿债压力。 要消化如此庞大的新增算力，本身就需要AI产业收入保持极高增速。报告援引当前估算称，OpenAI和Anthropic合计年收入约1000亿美元（约合6713亿元人民币）。如果以这一收入规模作为参照，要在2032年达到3.725万亿美元，需要未来数年保持大约80%的年复合增长率。 以英伟达为例，它一边向云厂商和数据中心出售GPU，一边又通过残值担保等方式帮助客户获得融资，降低大规模采购芯片的资金门槛。云厂商则继续建设数据中心、向模型公司出售算力，数据中心投资者再依靠大型科技企业的长期租赁和采购承诺支撑债务。 一旦需求增长赶不上数据中心和GPU产能释放，芯片价格、算力租赁价格和设备利用率都可能下降，而已经签下的债务、租金和采购承诺仍需继续支付。AI行业信号频道: 前沿科技本内容来源于网络 原文链接，观点仅代表作者本人，

匹配字段正文

or to what Gen-2 could produce. There were a lot of, a lot of chatter on Twitter about Runway. Runway’s done. like, there is no way Runway will catch up. And if you remember, also OpenAI in the early twenty-It felt very, like it’s a Swyx [00:36:56]: To the moon Anastasis [00:36:57]: It’s a formidable opponent now, but at that point, it, they were on the top of their game. nobody could even get close to them. There was maybe Gemini was just the first version of Gemini had just released. So when O

匹配字段正文