Skip to content
Source · Daily Brief

AI Daily Brief — 09 March 2025

Sunday turned into Manus’s first reality check. TechCrunch’s Kyle Wiggers published a hands-on review under the headline “probably isn’t China’s second DeepSeek moment” — testing four tasks (food order, flight booking, restaurant reservation, Naruto game) that all failed or crashed. The piece confirmed Manus is an agent framework on top of Anthropic Claude 3.5 and Alibaba Qwen, not a proprietary foundation model — distinguishing it from DeepSeek, which trained and open-sourced R1 / V3 in-house. Research lead Yichao “Peak” Ji had claimed superiority over OpenAI’s Deep Research and Operator. The invite-code secondary market on Xianyu kept boiling at the $7,000-$13,800 band. The UK Artificial Intelligence (Regulation) Bill — laid in the House of Lords on March 4 — was in its first-reading window over the weekend, proposing a central AI Authority and statutory principles for safety, accountability, and governance. NVIDIA’s GTC 2025 keynote sat nine days away.

Top stories

  • TechCrunch publishes skeptical Manus deep-dive. “Probably isn’t China’s second DeepSeek moment.” Kyle Wiggers tested four tasks (food order, flight booking, restaurant reservation, Naruto game) — all failed or crashed. Piece notes Manus orchestrates Claude 3.5 + Qwen, doesn’t develop its own models. via TechCrunch
  • Manus invite-code secondary market explodes. Three days after the March 6 invite-only launch, codes were reselling on Xianyu for ¥50K-100K ($7,000-$13,800). The official Discord crossed 138K members within days; the launch demo drew over one million views in twenty hours. via Wikipedia
  • Manus identified as Claude 3.5 + Qwen orchestrator, not a new model. Coverage clarified the distinction from DeepSeek (which trained and open-sourced R1/V3 in-house). Ji’s claimed superiority over OpenAI’s Deep Research and Operator drew skepticism after the failed TechCrunch tests. via TechCrunch
  • UK Artificial Intelligence (Regulation) Bill in first-reading window. Laid in the House of Lords March 4. Proposes a central AI Authority and statutory principles for safety, accountability, and governance. No parliamentary action on Sunday; debate to resume the following week. via Anecdotes AI

Who shipped

Nothing. OpenAI, Anthropic, Google DeepMind, Meta, xAI, and Mistral all silent on Sunday.

Open-source pulse

Alibaba’s QwQ-32B reasoning model (March 6 release) continued leading the weekend Chinese open-source conversation — 32B reasoning model Alibaba positioned as matching DeepSeek-R1’s performance. No new open-source releases on Sunday; community discussion on X centered on QwQ-32B benchmarks and the Manus-as-Qwen-wrapper revelation.

Quiet corners

arXiv announcement system paused for the weekend. NVIDIA’s GTC 2025 keynote was scheduled for March 18 — nine days out — and the broader chip / infra news cycle was paused ahead of GTC week. Alibaba’s prior-week 380-billion-yuan (~$52.4B) three-year AI and cloud commitment continued anchoring infra commentary.

By the numbers

  • 4 / 4 — failed TechCrunch hands-on Manus tasks
  • 9 days until NVIDIA GTC 2025
  • ~$52.4B / 3 years — Alibaba AI + cloud commitment frame
  • Most-mentioned company: Manus (skeptical)

Compiled by AI Feed’s editor from verified web sources for 9 March 2025.