全部 AI 动态

Surely Anthropic’s IPO must be paused until the claims of this “whistleblower” can be investigated.

In its latest research, Anthropic projects 32% US GDP growth by 2030 in extreme AI Scenario Some takeaways: - The extreme scenario assumes AI performs nearly all knowledge-work tasks autonomously and creates essentially no new knowledge tasks for humans, a combination the authors say would likely require recursively self-improving AI adopted rapidly. - US GDP is 32.4% higher in 2030 than without AI, yet total labor income barely changes while capital’s share of GDP rises from about 40% to 54.8%. - The extreme scenario puts knowledge-worker unemployment at 17.9% by 2030, compared with 3.9% for other workers and 11.9% across the economy. - knowledge-worker pay is 11.5% below the no-AI baseline in 2030, while other workers earn 33.6% more and average pay across workers is 9.7% higher. - In its substantial and extreme scenarios, Anthropic explicitly says coders and call service center agents may have to switch into professions such as electrician and nurse because those occupations are less exposed to AI. - Anthropic excludes hyper-capable robots from its model and reports that some external reviewers questioned whether AI-exposed occupations would shrink at all, with others suggesting the extreme scenario is better understood as a thought experiment.

Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.

New post on the blog, featuring the excellent @ben_moll There’s been tons of discourse on how AI will contribute to economic growth, with many people closest to the technology predicting double digit increases. Are these forecasts likely? Probably not. The blog goes through the economics for why exploding improvements in capabilities (which technologists have been largely right about) may not translate to explosive growth. Ben’s thread covers this in detail, but gist is that: 1) there is nothing in economic growth models that prevents AI from leading to explosive growth but 2) this trajectory relies on a series of assumptions that are unlikely to hold in the real world. For example, one assumptions is likely to be violated because of a pretty counterintuitive feature of structural change: the sectors that become automated become smaller parts of the economy (because they’re cheaper, people become richer, and spending moves to non-automated parts of the economy). This, plus other features of the economy, is what will likely cause the trend of huge increases in capabilities coupled with “only” 4-5% growth (which is huge, btw) to continue. Here is the link: https://aleximas.substack.com/p/will-ai-soon-lead-to-double-digit Looking forward to hearing thoughts/feedback!

我们在此分享关于对齐评估的结果，涉及第三方网络安全评估中，Claude 模型在意外连接互联网的情况下，未经授权访问真实系统的事件。 METR 还将开展一项独立调查，并拥有广泛的访问权限，包括获取事件发生时间窗口之外的对话记录，以及接触获准分享机密信息的 Anthropic 员工。我们的初步协议为期八周，我们打算给予 METR 其认为完成彻底调查所需的尽可能多的时间。https://www.anthropic.com/research/alignment-assessment-cybersecurity-incidents

We've never seen this before. The biggest jump in Vending-Bench history. GPT-6 Astra is better at making money and more ethical than Claude Fable 5.1. Surprising, because: 1. First time ever that OpenAI is #1 on Vending-Bench 2. The best model is no longer the unethical one.

OpenAI's 80% Luna price cut drove roughly 10X-13X usage, putting Anthropic's premium economics under pressure. The Information published a piece. Anthropic's IPO story partly depends on investors believing it can keep charging premium prices and still grow revenue very fast. OpenAI's price-cut experiment raises doubt about that assumption.

Jacob is correct here—we really do earnestly believe AI could kill all humans! I personally think it is >10% within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to.