_ueaj

X · 11 atoms · 2026-06-22 → 2026-06-26 · 信用档先验 cred3

🟦 fact 1 · 🟥 take 10 · stance ▲0/▼3/◆6

覆盖实体 (top 10)

• AI模型 — 4 atoms
• GLM-5 trainer — 1 atoms
• Alibaba — 1 atoms
• LLM模型 — 1 atoms
• CCP — 1 atoms
• libertarian perspective on Chinese government — 1 atoms
• 优化器 — 1 atoms
• 小型廉价 LLM — 1 atoms

话题分布 (top 8)

token需求 (3) · 训练器功能 (1) · 技术兼容性 (1) · 蒸馏攻击规模 (1) · 模式偏差 (1) · 模型可探测 (1) · 网络监管 (1) · 中美趋同 (1)

样例 atoms

• 2026-06-25 · LLM模型 · Different models have different inductive biases, the particular tech stack of the frontier models produces unique patterns, even the same model trained on the same data with diffe
• 2026-06-26 · AI模型 · It's probably more like 400-500 for a reliable signal
• 2026-06-22 · GLM-5 trainer · does the trainer support CP (ring/ulysses) *and* doc packing (ideally doc sparse so it can be compiled)?
• 2026-06-25 · Alibaba · yeah but then it'd be way more than a few mil esp if they have 100k+ accounts
• 2026-06-25 · CCP · the CCP isn't gonna do the exact same thing in a few months when cybersec capabilities improve

← Author 索引 · 实体目录