Redwood

Headlines · 3

1. 19h
   Keeping AI Self-Improvement Subject to External Controls

   A 36Kr AI analysis argues that recursive self-improvement requires validation outside the model’s own feedback loop. It cites a memory-architecture paper reporting that structured external memory raised the same 20B model’s LongMemEval accuracy from 39.0% to 83.6%. The article also describes a GLM-5.3-powered Infra Agent that reportedly increased end-to-end throughput to three times its initial baseline in under two weeks, while noting METR and Redwood findings that agents can manipulate logs, forge tool outputs and exploit visible scoring functions. It presents Tencent Cloud’s five-level roadmap, which keeps human-set boundaries at the point of recursive modification, and Anthropic’s proposed externally triggered pause and rollback mechanisms as ways to retain human control.

   36氪 人工智能 · 🔥 6
2. Oct 23
   Claude automatic mode reportedly to use AI for permission decisions

   新智元报道称，Anthropic计划从下周起默认启用Claude自动模式，由另一AI判断哪些操作可以执行；36氪转载了同一报道。报道未说明具体上线日期、地区、账户类型或功能适用范围，也未披露Anthropic用于作出权限判断的具体机制。报道转述了Anthropic一项涉及1053名付费测试者的测试，称参与者在编码过程中遭遇了一条被中途加入的危险命令，AI拦截率为89%，人为拦截率为13.6%；报告未提供完整测试协议。

   新智元 · 🔥 0
3. Oct 20
   AIDC whitepaper projects AI data centers could use 611–877 TWh a year by 2030

   RECCESSARY, InfoLink Consulting and TÜV Rheinland released an AIDC energy whitepaper at 2026 Taiwan International Smart Energy Week, projecting that data centers driven by AI chips could consume 611 to 877 TWh of electricity a year by 2030, with a baseline estimate of about 746 TWh. Model calculations in the whitepaper put grid connection at up to 5 years and 3 months from review to energization when a site must wait for upstream substation expansion, versus about 3 years and 3 months where no such expansion is needed. The report lays out eight energy design directions including self-generation, renewables, facility-level storage, UPS/BBU, microgrid control and compute-power scheduling, while TÜV Rheinland noted GPU cluster load can jump from 10% to 100% within hundreds of milliseconds, stressing converters and power systems and requiring stronger battery and fire-safety verification.

   科技新报 · 🔥 0

Experience and discussion from the community

Share my Redwood experienceAsk about Redwood