以下部分引用 AI 总结，现阶段 AI 仍然有幻觉。请以内容的原文为准。
数据新鲜度群聊 07-14 ✓卖方 07-14 ✓Wrap 断档15天 ✕News 07-15 ✓X 断档12天 ✕页面生成 07-15
RSI
15 atoms · 跨 10 天 · 首见 2026-04-23 · 最近 2026-06-29
三色: 🟦 fact 0 · 🟥 take 15 · stance ▲9/▼3/◆1
来源: X 15
时态: fresh:11 · aging:4
标签: 好观点:8 · 好思考:6 · 好问题:1
叙事 (narrative) (12)
- 2026-06-29 · 解决内存瓶颈可能性
what if RSI solves the memory bottleneck
tags: 好思考 · → daily
- 2026-06-28 · 发生迹象
hints of RSI happening in these systems
tags: 好思考 · → daily
- 2026-06-25 · definition debate
You can define RSI in a way where it’s been happening for years: “humans/AI work together to make better AI” or in a way where it hasn’t happened yet: “AI, by itself, makes better AI”
tags: 好思考 · → daily
- 2026-06-07 · 递归效应（RSI）对历史技术突破是独特的
Recursion (and thus RSI) effects would clearly be unique relative to historical technology breakthroughs.
tags: 好思考 · → daily
- 2026-05-19 · 递归自我改进可能改变竞争动态
RSI - recursive self-improvement - could change the competitive dynamics.
tags: 好思考 · → daily
- 2026-04-23
[老化中] · 状态
I am calling it rsi is in full swing
tags: 好思考 · → daily
value: direction=up
- 2026-06-21 · 大杠杆在evals
the big levers are evals, data, and kernels. but data and kernels are evals problems, so it’s really just evals
tags: 好观点 · → daily
- 2026-06-21 · 大杠杆在bring GPUs online
that, and bringing the damn GPUs online
tags: 好观点 · → daily
- 2026-06-10 · RSI转化为通用智能的程度不确定
this is a series of questions - how much your ML/SWE RSI translates to general intelligence - how much can int accelerate engineering - how much in our industrial processes can be skipped via better engineering if 200 IQ claude just writes a nanoassembler spec, welp
tags: 好问题 · → daily
- 2026-06-07 · mutual conditional pause agreement sentiment
now on the eve of RSI it seems everyone is more mutual conditional pause agreement pilled than they used to be and that seems like a good development
tags: 好观点 · → daily
- 2026-06-07 · risk vector assessment
I continue to think that RSI is an overrated risk vector due to data and compute bottlenecks, and that slowing down AI would accomplish little at enormous cost.
tags: 好观点 · → daily
- 2026-05-03
[老化中] · practical manifestation in AI labs
RSI, like many other pieces of LessWrong lore, is poisoned by capeshit imagination. This is how it looks in reality. Not models rewriting their weights directly by means of perfect mechanistic self-interpretability, but discovering better recipes and data.
tags: 好观点 · → daily
预测 (forecast) (1)
- 2026-06-08 · timeline for machine learning/CS
that experience alone lengthens my timelines for RSI by a year, at least for machine learning / CS
tags: 好观点 · → daily
value: date=+1 year
立场 (position) (2)
- 2026-06-25
[老化中] · coming
I am more convinced than ever that RSI is not coming.
tags: 好观点 · → daily
- 2026-06-25
[老化中] · is already happening
RSI is already happening.
tags: 好观点 · → daily
时间轴 (近 20)
- 2026-06-29 ·
narrative · 解决内存瓶颈可能性 · → daily
- 2026-06-28 ·
narrative · 发生迹象 · → daily
- 2026-06-25 ·
position · coming · → daily
- 2026-06-25 ·
position · is already happening · → daily
- 2026-06-25 ·
narrative · definition debate · → daily
- 2026-06-21 ·
narrative · 大杠杆在evals · → daily
- 2026-06-21 ·
narrative · 大杠杆在bring GPUs online · → daily
- 2026-06-10 ·
narrative · RSI转化为通用智能的程度不确定 · → daily
- 2026-06-08 ·
forecast · timeline for machine learning/CS · → daily
- 2026-06-07 ·
narrative · 递归效应（RSI）对历史技术突破是独特的 · → daily
- 2026-06-07 ·
narrative · mutual conditional pause agreement sentiment · → daily
- 2026-06-07 ·
narrative · risk vector assessment · → daily
- 2026-05-19 ·
narrative · 递归自我改进可能改变竞争动态 · → daily
- 2026-05-03 ·
narrative · practical manifestation in AI labs · → daily
- 2026-04-23 ·
narrative · 状态 · → daily
← 实体目录 · 系统日志