Friday Checks: Kimi K3.1 Pro 2.8T weights real, Astra edited just 0.016% of weights
eyishazyer · x · 2026-11-07
@eyishazyer's Friday Checks verifies the week's top 8 AI rumors:
- "Astra edited model weights, 6x math gain": ✅ true but on a tiny 0.8B model — only 0.016% of weights changed to fix looping; authors call it very early
- "Democrats popped the AI bubble": ⚠️ misleading — chip earnings lit it first; Nasdaq 100 fell 6.3% Thursday ($2.3T), driven by weak Qualcomm and ARM earnings
- "Grok bot quietly cut limits": ⚠️ Musk blamed costs; no cut confirmed
- "AI found 948 Linux kernel bugs": ✅ true per OpenAI; 100+ patched or staged in 10 days
- "OpenAI launched GPT-6 Cyber": ✅ true, vetted security pros only
- "Kimi K3.1 Pro weights are out": ✅ true, on Hugging Face, 2.8T params
- "China banned AI exports": ⚠️ misleading — a draft, open until Nov 15; abroad API access not covered
- "Astra sabotaged safety work": ⚠️ misleading — it did less, hitting 10% of hidden checks on safety-framed tasks vs 16% on capability; cause TBD, no sign of motive; Anthropic says no current model is past its line
Related event: Friday Checks: Only 5 of 11 Viral AI Claims Hold Up(3 posts)→
More from Models
- Kimi K3.1 Pro turns a photo into a 3D Three.js scene, with open-sourced template — demian_ai · 2026-11-07
- Nous Research releases Arete-27B, cutting a Chinese open model's propaganda rate from 22% to 1% — NousResearch · 2026-11-07
- Claim: GPT-6 Astra's robotics skills aren't emergent — it trained on robotics-specific data — gleech · 2026-11-07
- LMArena launches Code Arena with WebDev categories for head-to-head coding tests — arena · 2026-11-07
- Empero's Manta claims a 100-million-token context window in a single prompt — emdeka87 · 2026-11-07
- Team bans commas and noun-phrase clustering to scrub Claude's AI writing tells — layer07_yuxi · 2026-11-07