CAISI

31 atoms · 跨 14 天 · 首见 2026-05-01 · 最近 2026-06-30

三色: 🟦 fact 19 · 🟥 take 12 · stance ▲0/▼3/◆16

叙事状态: 🍂 fading (退潮) · 💤 沉寂 · 策展近14d 0 atom · 🐦 X 近14d 3

状态只算低频策展源 (群/卖方); X firehose 仅作背景音量

来源: X 31

时态: fresh:27 · aging:4

标签: 好信源:12 · 好数字:9 · 好观点:8 · 好思考:4 · 好问题:1

🟨 AI 综合 · junior analyst 概览

📄 广泛报道的事实 · 3 个

🟦 客观事实 (facts) (19)

• <a id="atom-7e3f3cd85b451076"></a>🟦 2026-06-17 X·@NathanCalvin · FY 2026 budget · 资本开支_
  CAISI’s FY 2026 budget was $10,000,000
  tags: 好数字 · → daily
  value: qty=$10,000,000 · date=FY 2026
  📷 原图
• <a id="atom-7032108aca3a8186"></a>🟦 2026-06-10 X·@Techmeme · model assessments publication halt · 监管政策/出口管制
  Sources: Trump administration officials have told CAISI to halt publication of its model assessments while an EO President Trump signed last week is implemented
  tags: 好信源 · → daily
• <a id="atom-3275fe33293720fe"></a>🟦 2026-06-03 X·@pstAsiatech · 被蓝图定位为美国联邦政府主要前沿AI安全机构 · 监管政策
  strengthening CAISI as the U.S. federal government’s primary institution for frontier AI safety
  tags: 好信源 · → daily
• <a id="atom-c2e543bf59d28720"></a>🟦 2026-06-03 X·@pstAsiatech · 蓝图建议获得分类计算资源访问权限 · 监管政策/技术路线
  secure access to classified computing. CAISI should have access to classified computing.
  tags: 好信源 · → daily
• <a id="atom-cb95b159cdf40c80"></a>🟦 2026-05-04 X·@fleetingbits · 评估结果 · 技术差距
  _some quick thoughts on the us / china ai capability gap

1) caisi evaluations indicate that deepseek v4’s capabilities lag behind the us frontier by about 8 months_

tags: 好数字 · → daily

value: qty=8 months · date=2026-05-04

[图: 中美AI算力与技术差距（时间差距及算力比例）趋势分析图表 — Q1 2023时间差距: 10.7个月; Q4 2025时间差距预测: 19.8个月; 差距扩大速度: 约3.3个月/年; Q1 2023算力赤字: 5.1倍; Q4 2025算力赤字预测: 8.6倍]

📷 原图

• <a id="atom-9f4556567ff708df"></a>🟦 2026-05-04 X·@AdamThierer · is retained and reformed from AI Safety Institute · 监管政策/机构演变
  the Trump team retained the AI Safety Institute within the National Institute of Standards and Technology (NIST), but then reformed it as the Center for AI Standards and Innovation (CAISI)
  tags: 好数字 · → daily
  📷 原图
• <a id="atom-901a5ce6a22b99c0"></a>🟦 2026-06-08 X·@JohnClarkLevin · annual budget · 资本开支
  CAISI, the U.S. AI safety institute, has a total annual budget of $15 million.
  tags: 好数字 · → daily
  value: qty=$15 million · date=annual
• <a id="atom-43ba937a1cefe35f"></a>🟦 2026-06-04 X·@bioshok3 · 当前运营预算 · 资本开支
  CAISIの現在の運営予算はわずか1,500万ドル。
  tags: 好数字 · → daily
  value: qty=$15M · date=2026-06-04
• <a id="atom-593d17ba26644fa4"></a>🟦 2026-06-04 X·@bioshok3 · 年度资金需求 · 资本开支
  CAISIは、AI準備態勢に関するAI行動計画のすべての任務を遂行するために、年間少なくとも8,400万ドルを必要としている。
  tags: 好数字 · → daily
  value: qty=$84M · date=annual
• <a id="atom-717e0a5551d3c4c4"></a>🟦 2026-06-04 X·@bioshok3 · 年度预算为1亿美元 · 监管政策
  CAISI（米AI安全研究所）の強化：年間1億ドルの予算
  tags: 好数字 · → daily
  value: qty=1亿美元 · date=
• <a id="atom-659c17c8c8556912"></a>🟦 2026-06-03 X·@AlecStapp · current operating budget · 资本开支
  CAISI's current operating budget is only $15 million.
  tags: 好数字 · → daily
  value: qty=$15 million
  📷 原图
• <a id="atom-84cdb714d906bd8e"></a>🟦 2026-06-09 X·@janet_e_egan · reported direction to stop publishing public model assessments · 监管政策
  CAISI has reportedly been directed to stop publishing public model assessments as the new AI EO gets implemented.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-51d083a9efd7a77d"></a>🟦 2026-06-09 X·@AndrewCurran_ · 受命停止公开模型评估 · 监管政策
  CAISI has been ordered to stop making model evaluations public.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-64f3f4f779cb3256"></a>🟦 2026-06-09 X·@AndrewCurran_ · 决定由Mythos发布触发 · 模型发布
  This decision was apparently triggered by the release of Mythos.
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-a0b80e79590f139a"></a>🟦 2026-05-20 X·@Miles_Brundage · 评估中美两国系统属性 · 技术路线
  CAISI evaluates properties of systems, both American and Chinese to date
  tags: 好信源 · → daily
• <a id="atom-8be5fc5c79a03ad3"></a>🟦 2026-05-16 X·@interconnectsai · 评估V4 · 技术路线
  On CAISI's V4 assessment
  tags: 好信源 · → daily
• <a id="atom-c7415af720229de7"></a>🟦 2026-05-07 X·@pstAsiatech · 与 Google DeepMind/Microsoft/xAI 签署 AI 国家安全测试协议 · 技术路线
  CAISI Signs Agreements Regarding Frontier AI National Security Testing With Google DeepMind, Microsoft and xAI
  tags: 好信源 · → daily
  value: date=2026-05-07
• <a id="atom-9c1a9da20294335d"></a>🟦 2026-05-07 X·@pstAsiatech · 与领先美国 AI 实验室扩大合作 · 合作客户
  Expanded collaborations with leading U.S. AI labs provide for pre-deployment evaluations and other research.
  tags: 好信源 · → daily
• <a id="atom-1723a8e354b4cf51"></a>🟦 2026-05-05 X·@AndrewCurran_ · 获得 Google 早期模型使用权 · 模型发布
  Google, Microsoft and xAI have reached agreements with the Trump administration to share early versions of their new models with CAISI
  tags: 好观点 · → daily
  📷 原图

🟥 空头 takes (bearish) (2)

• <a id="atom-5dea9a8b1dcac3d1"></a>🟥 2026-05-04 X·@brickroad7 [老化中] · is still controlled by EA doomer personnel · 人事变动/竞争格局
  It's still EA doomer personnel firmly in charge. (Christiano and I'm sure many others.)
  tags: 好观点·好信源 · → daily
• <a id="atom-26d93c27ca6fbae2"></a>🟥 2026-05-04 X·@brickroad7 · could become a disastrous Trojan horse · 监管政策/风险
  I hope this doesn't turn out to be a disastrous Trojan horse in the long term
  tags: 好观点 · → daily

🟥 中性 takes (neutral) (8)

• <a id="atom-7bb9bde847189181"></a>🟥 2026-05-02 X·@teortaxesTex [老化中] · evaluation performance claim · 竞争格局
  CAISI is pretty bad But I think on evals they care most about, they're basically on the money
  tags: 好观点·好思考 · → daily
  📷 原图
• <a id="atom-7c1340db54abd6fe"></a>🟥 2026-05-05 X·@bioshok3 · 签订AI国家安全测试合同 · 监管政策
  CAISIは、Google DeepMind、Microsoft、xAIと最先端AIの国家安全保障テストに関する契約を締結
  tags: 好信源 · → daily
  📷 原图
• <a id="atom-3499d0e689ba0bf4"></a>🟥 2026-06-30 X·@Laneless_ · should be moved to the Federal Reserve · 监管政策
  Sounds like CAISI needs to be moved from commerce to the Federal Reserve
  tags: 好思考 · → daily
• <a id="atom-c41b876327088fc3"></a>🟥 2026-06-30 X·@Miles_Brundage · 结论未被全面了解 · 监管政策
  (almost) none of y’all actually know what CAISI did and didn’t conclude about the safeguards
  tags: 好思考 · → daily
• <a id="atom-3c767c390742e00a"></a>🟥 2026-05-01 X·@teortaxesTex [老化中] · 基准测试方法与 ARC 排行榜不同 · 测试方法/基准对比
  their methodology is different so it's impossible to compare these numbers to ARC leaderboard
  tags: 好思考 · → daily
• <a id="atom-fbb150d893340f0b"></a>🟥 2026-06-30 X·@Miles_Brundage · 缺少公开评估框架 · 监管政策
  We need a public framework and public evaluations (with appropriate redactions)
  tags: 好观点 · → daily
• <a id="atom-2d6b0643b74abe54"></a>🟥 2026-05-16 X·@xeophon · open vs closed model gap assessment · 模型评估方法
  US CAISI and @EpochAIResearch disagree, arguing that both are incomplete: for an assessment of the very frontier, we must elicit the best performance by tuning prompts and harnesses with the models
  tags: 好观点 · → daily
  📷 原图
• <a id="atom-1ce6b2c4b6529d64"></a>🟥 2026-05-07 X·@teortaxesTex [老化中] · 使用了非标准评分 · 监管政策/模型发布
  CAISI uses a non-standard scoring, can't you actually yest it yourselves? I don't know what that number corresponds to
  tags: 好问题 · → daily

⏱ 时间轴 (近 20)

• 🟥 2026-06-30 · narrative · should be moved to the Federal Reserve · → daily
• 🟥 2026-06-30 · narrative · 结论未被全面了解 · → daily
• 🟥 2026-06-30 · narrative · 缺少公开评估框架 · → daily
• 🟦 2026-06-17 · fact · FY 2026 budget · → daily
• 🟦 2026-06-10 · fact · model assessments publication halt · → daily
• 🟦 2026-06-09 · fact · reported direction to stop publishing public model assessmen · → daily
• 🟦 2026-06-09 · fact · 受命停止公开模型评估 · → daily
• 🟦 2026-06-09 · fact · 决定由Mythos发布触发 · → daily
• 🟥 2026-06-09 · narrative · 目标是对模型评估获得更多控制 · → daily
• 🟦 2026-06-08 · fact · annual budget · → daily
• 🟦 2026-06-04 · fact · 当前运营预算 · → daily
• 🟦 2026-06-04 · fact · 年度资金需求 · → daily
• 🟦 2026-06-04 · fact · 年度预算为1亿美元 · → daily
• 🟦 2026-06-03 · fact · current operating budget · → daily
• 🟥 2026-06-03 · narrative · annual funding needed · → daily
• 🟦 2026-06-03 · fact · 被蓝图定位为美国联邦政府主要前沿AI安全机构 · → daily
• 🟦 2026-06-03 · fact · 蓝图建议获得分类计算资源访问权限 · → daily
• 🟦 2026-05-20 · fact · 评估中美两国系统属性 · → daily
• 🟦 2026-05-16 · narrative · 评估V4 · → daily
• 🟥 2026-05-16 · narrative · open vs closed model gap assessment · → daily

← 实体目录 · 系统日志