字节跳动等机构发布75页论文,将AI自优化分为5级,当前多数模型处于低级阶段
朋友,字节跳动和清华等团队搞了个75页的论文,把AI自己优化升级的能力分了五级,现在看主流模型都还处在比较初级的阶段,挺有意思的。
这篇由字节跳动、清华大学等机构联合发布的75页论文,将AI自优化能力分为五个等级。在Level 1阶段,人类仍决定AI的改进方向;
而Level 5则意味着AI能自主重写构建其后续版本的过程。论文对491篇现有论文进行分类,结果显示75%的模型处于Level 1或2,仅有不到6%达到Level 5。
OpenAI、Anthropic和Google DeepMind等西方实验室已将AI自动化研发列为安全框架中的风险。
"Everything AI has achieved so far is but a drop in the ocean." That's page one of a 75-page paper called "The Last AI Built by Humans," from 30+ researchers at ByteDance, Tsinghua, Shanghai AI Lab and six other institutions. The paper ranks AI self-improvement on five levels. At Level 1, humans still decide what gets fixed. By Level 5, the AI is rewriting the process that builds its own successor. They also sorted 491 existing papers onto the ladder, and 75% landed at Level 1 or 2. Under 6% made it to Level 5. Level 5 is what Western labs have been warning about. OpenAI, Anthropic, and Google DeepMind each list AI automating its own R&D as a risk in their safety frameworks.📊 4 ⚡
本条情报归属 AITOP「论文」核心栏目,重点聚焦中国 AI 产业与国产大模型生态。在跨区域产业观察坐标中,该动态直接关联「各国 AI 监管法案、合规治理与主权算力政策」「中美大模型技术代差与商业化对照」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:字节跳动、AI自优化、Level 5、OpenAI,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。
信源核验说明:本文原始线索采编自「The Rundown AI」(@therundownai),经 AITOP 多维评分管线结构化提炼。截至目前,全网已有 11 家独立信源对该事件进行交叉印证与跟进报道(含 AI Will、arXiv: Google DeepMind、theverge 等),属于近期高关注度行业信号。