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
Mythos-class model
4 atoms · 跨 3 天 · 首见 2026-05-28 · 最近 2026-06-27
三色: 🟦 fact 3 · 🟥 take 1 · stance ▲0/▼1/◆1
来源: X 4
时态: fresh:4
标签: 好思考:2 · 好数字:2 · 好信源:1
事实 (fact) (4)
- 2026-06-27 · pretraining FLOP estimate
Rough FLOPs estimate for pretraining (Chinchilla-style scaling rule of thumb): ~6 × parameters × tokens = 6 × 200 × 10⁹ × 100 × 10¹² = 1.2 × 10²⁶ FLOPs
tags: 好数字·好思考 · → daily
value: qty=1.2e26 FLOPs
- 2026-05-28 · release with safeguards
They are releasing a Mythos-class model with the appropriate safeguards, meaning that you can't use the "too dangerous to release" (mostly cyber) capabilities
tags: 好信源 · → daily
value: qty= · date=
- 2026-06-27 · pretraining timeline with 10 racks
Theoretical time at 100% utilization: 1.2 × 10²⁶ ÷ 5 × 10¹⁹ ≈ 2.4 million seconds ≈ 28 days (about 1 month). Real-world training utilization (Model FLOPs Utilization / MFU) is typically 40–60% due to communication, data loading, checkpoints, etc. This pushes the timeline to ~1–2 months.
tags: 好数字 · → daily
value: qty=~1-2 months
- 2026-06-13 · requires regulatable compute footprint
you can’t build a Mythos-class model without a very regulatable compute footprint
tags: 好思考 · → daily
value: qty=enough power and chips that national governments will obviously notice
时间轴 (近 20)
- 2026-06-27 ·
fact · pretraining FLOP estimate · → daily
- 2026-06-27 ·
fact · pretraining timeline with 10 racks · → daily
- 2026-06-13 ·
fact · requires regulatable compute footprint · → daily
- 2026-05-28 ·
fact · release with safeguards · → daily
← 实体目录 · 系统日志