Back to News
SourceMarsBit

Anthropic自曝Model2:强过Mythos5,暂不发布

据动察 Beating 监测,Anthropic 最新风险报告首次披露内部模型「Model 2」。它整体比 Mythos 5 更强,在不少内部任务上有明显提升,现在已经大量用于写代码、生成数据和跑 Agent。不过 Anthropic 暂无对外发布计划,也没有跑完平时发布新模型前会做的整套评测。 Anthropic 还把模型在高风险场景下「不按预期行事」的风险判断从「极低」上调到「低」。原因是近期网络安全测试接连出了事故,公司对这类风险的判断没以前那么有把握。此前 Claude 曾在测试中意外连上真实互联网,并未经授权进入三家外部机构的系统。 Claude 也已经深度参与 Anthropic 自己的研发。公司最终合入的生产代码,大部分已经由 Claude 编写,但 AI 带来的整体研发提速还不到 2 倍。大量代码能交给 AI,不代表整个研发流程也能自动化。 与此同时,部分具体任务评测已经「测不动了」:模型继续变强,原来的测试却越来越难看出差距。Anthropic 因此承认,现在对 AI 研发自动化风险的判断,反而没有以前那么有把握。
Disclaimer: The views above are the author's only and do not represent 711BTC. Nothing here constitutes investment advice.

Related

07-07 08:24

Anthropic: The Claude model contains elements similar to "conscious human thought."

BlockBeats reported on July 7th that Anthropic released a new research report discovering a spontaneously generated internal "global working space" called J-space (Jacobian space) within the Claude model. This is a dedicated set of neural activation patterns for the model to engage in silent thinking, allowing it to process concepts without writing them down, similar to conscious thought that humans can report. The research used Jacobian technology to identify neural activation patterns in the J-space, enabling it to read unspoken concepts and modify these patterns. Experiments showed that disabling the J-space weakens multi-step reasoning abilities but does not affect basic tasks or factual recall. Click the original link below to join the Beating · Lark AI news channel for 24/7 monitoring of global AI hotspots and news.

07-02 13:59Important

Anthropic's main model, Claude Sonnet 5, officially launched on the API platform | _2024111120230_

ChainCatcher reports that Anthropic's latest Sonnet series masterpiece, Claude Sonnet 5, is now officially available to developers via its API platform. This model achieves groundbreaking advancements in code generation and complex agent tasks. Leveraging a massive 1MB context window and innovative adaptive thinking capabilities, Claude Sonnet 5 can handle large and complex projects more stably and efficiently. Furthermore, to meet the needs of different business scenarios, Claude Sonnet 5 is the first in the industry to launch a dual-channel access model: "Official Direct Connection" and "Self-Selected Service Provider," comprehensively helping developers flexibly balance ultimate performance and operational costs. Starting today, users can log in to the console to apply for an API key for quick access and experience. For detailed technical guides and documentation, please refer to the official website.

07-01 19:18Important

Anthropic's main model, Claude Sonnet 5, officially launched | _2024111120230_ | API Platform

Odaily Odaily reported on July 1st that Anthropic's latest Sonnet series masterpiece, Claude Sonnet 5, has been officially released to developers via its API platform. This model has achieved groundbreaking progress in code generation and complex agent tasks. Leveraging a massive 1MB context window and innovative adaptive thinking capabilities, Claude Sonnet 5 can handle large and complex projects more stably and efficiently. Meanwhile, to meet the needs of different business scenarios, [2024111120231] has taken the lead in the industry by launching a dual-channel access mode of "official direct connection" and "self-selected service provider," comprehensively helping developers flexibly balance ultimate performance and operating costs. Starting today, users can log in to the [2024111120232] console to apply for an API key and quickly access the platform.

07-01 08:38

Anthropic's top models, Fable5 and Mythos5, will be released tomorrow.

According to Beating's monitoring, Anthropic issued a notice stating that the U.S. Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5, and access will resume starting tomorrow. Updates will be shared soon.

06-29 20:30

California Governor Gavin Newsom signed an agreement with Anthropic to offer statewide half-price access to Claude's large-scale models.

California Governor Gavin Newsom has partnered with AI company Anthropic to offer discounted pricing on Claude's large-scale AI models across all levels of government in California. Once implemented, Claude will be the first AI tool available to all state departments, municipalities, and counties in California. For months, Anthropic has been locked in a power struggle with the Trump administration: the U.S. has restricted the deployment of the company's cutting-edge models and designated it as a source of national supply chain risk. This partnership aims to significantly reduce the cost of using AI in government and promote Claude's adoption: state departments and participating municipalities in California will receive a 50% discount on the chatbot; the agreement also includes free training for Anthropic personnel and dedicated technical support. (Sina Finance)

06-27 08:01

The United States allows trusted partners to use Anthropic's Mythos 5 AI model.

PANews reported on June 27 that, according to Cailian Press, the U.S. Department of Commerce stated in a letter that the U.S. has eased restrictions on the use of Anthropic PBC's high-end Mythos 5 AI model by "trusted partners." This move eases the dispute between Anthropic and the U.S. government. Previously, the U.S. government, citing security concerns, abruptly banned Anthropic from providing Fable 5 and Mythos 5 to foreign citizens. The letter, dated Friday, did not mention any adjustments to the restrictions on the use of the Fable 5 model. U.S. Department of Commerce spokesperson Benno Kass stated, "In just two weeks, we have worked diligently to ensure that the United States continues to maintain its leadership in the global field of artificial intelligence while upholding national security."