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
DeepSeek R1
15 atoms · 跨 12 天 · 首见 2026-04-25 · 最近 2026-07-02
三色: 🟦 fact 8 · 🟥 take 7 · stance ▲4/▼1/◆4
来源: X 15
时态: fresh:14 · aging:1
标签: 好观点:6 · 好数字:5 · 好信源:5 · 好思考:2
叙事 (narrative) (4)
- 2026-06-21 · 推理为涌现现象
reasoning does emerge on its own as an emergent phenomenon
tags: 好思考 · → daily
- 2026-06-19 · impact on AI industry perception
jan 2025, when r1 violently shook everyone awake
tags: 好观点 · → daily
- 2026-07-02 · 模型行为正常
DeepSeek R1 不必 mental illness
tags: 好观点 · → daily
- 2026-04-25
[老化中] · can make V2 relevant again
and R1 can make V2 relevant again.
tags: 好观点 · → daily
反驳 (refute) (1)
- 2026-05-01 · 参数数量估值修正
DeepSeek R1 (true size 671B): 424B -> 760B
tags: 好数字·好思考 · → daily
value: qty=760B
事实 (fact) (10)
- 2026-06-25 · token efficient
The model is less token efficient than 5.5
tags: 好观点·好信源 · → daily
value: qty=less · object=5.5
- 2026-06-04 · established multi-teacher SFT to RL standard
The multi-teacher SFT to RL that Microsoft did in their first model was the standard established by DeepSeek R1.
tags: 好信源·好观点 · → daily
- 2026-07-02 · reported language mixing, gibberish, and poor readability in the COT
DeepSeek R1 reported language mixing, gibberish, and poor readability in the COT
tags: 好信源 · → daily
- 2026-06-21 · 发布时间早于OpenAI共享推理轨迹
when DeepSeek R1 was released, at that time OpenAI did not share reasoning traces
tags: 好信源 · → daily
- 2026-06-06 · 月度推理成本
DeepSeek R1: ~$2,740
tags: 好数字 · → daily
value: qty=$2,740 · date=每月
- 2026-06-02 · 开源状态
It's Open Sourced if it wasn't clear
tags: 好信源 · → daily
value: direction=是
- 2026-05-27 · 初始定价
R1 launched at 2.19 usd per 1M
tags: 好数字 · → daily
value: qty=2.19 usd per 1M
- 2026-05-02 · Elo 评分
DeepSeek R1 Elo: 约180
tags: 好数字 · → daily
value: qty=180
- 2026-04-25 · RL训练成本
R1's RL took $82 K
tags: 好数字 · → daily
value: qty=$82K
- 2026-06-26 · 不是当前前沿但重要且独特、为下一步下载目标
Next I need to get deepseek R1 and L405. Not currently frontier but important and beautifully unique models.
tags: 好观点 · → daily
时间轴 (近 20)
- 2026-07-02 ·
narrative · 模型行为正常 · → daily
- 2026-07-02 ·
fact · reported language mixing, gibberish, and poor readability in · → daily
- 2026-06-26 ·
fact · 不是当前前沿但重要且独特、为下一步下载目标 · → daily
- 2026-06-25 ·
fact · token efficient · → daily
- 2026-06-21 ·
fact · 发布时间早于OpenAI共享推理轨迹 · → daily
- 2026-06-21 ·
narrative · 推理为涌现现象 · → daily
- 2026-06-19 ·
narrative · impact on AI industry perception · → daily
- 2026-06-06 ·
fact · 月度推理成本 · → daily
- 2026-06-04 ·
fact · established multi-teacher SFT to RL standard · → daily
- 2026-06-02 ·
fact · 开源状态 · → daily
- 2026-05-27 ·
fact · 初始定价 · → daily
- 2026-05-02 ·
fact · Elo 评分 · → daily
- 2026-05-01 ·
refute · 参数数量估值修正 · → daily
- 2026-04-25 ·
fact · RL训练成本 · → daily
- 2026-04-25 ·
narrative · can make V2 relevant again · → daily
← 实体目录 · 系统日志