Anthropic's "When AI builds itself": when AI starts building itself
Original title:Anthropic万字长文:当AI开始构建自己,人类该何去何从?
AI summary
Anthropic noted in a post that AI systems are accelerating their own development, and recursive self-improvement may be approaching. Currently more than 80% of Anthropic's code is written by Claude, and engineers merge 8 times as much code per day as in 2024. Claude Opus 4.6 can handle 12-hour software tasks, while Mythos Preview works continuously for at least 16 hours; benchmarks such as SWE-bench and CORE-Bench are saturated. Internal tests show Mythos Preview improves research efficiency by about 4 times, achieves 52x code speed optimization, reaches a 76% success rate on open-ended tasks, and closes 97% of the research project gap. Claude gives better research advice than humans in 64% of cases.
Key Takeaways
- Anthropic noted in a post that AI systems are accelerating their own development,and recursive…
- Currently more than 80% of Anthropic's code is written by Claude,and engineers merge 8 times as…
- Claude Opus 4.
- 6 can handle 12-hour software tasks,while Mythos Preview works continuously for at least 16 hours
- benchmarks such as SWE-bench and CORE-Bench are saturated.
Quick Facts
- Published
- Updated
- Source
- 公众号:数字生命卡兹克
- Category
- Opinion
- AI score
- 68
Anthropic首次公开内部数据,工程师代码产出提升8倍,Claude自我加速的曲线已经画出来了。这篇文章不是预测是事实,做AI的人都该花20分钟读完。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Anthropic's "When AI builds itself": when AI starts building itself. June 5, 2026. https://new.nat6.net/en/news/cmq2lrvf800kzsl6nvajb0wbi/