SemiAnalysis Slams NVIDIA: Committee-Based Frontier Model Development Does Not Work
teortaxesTex · x · 2026-08-13
SemiAnalysis commented on the release of DeepSeek v4 Pro, using the opportunity to heavily criticize NVIDIA's Nemotron3 Ultra.
They pointed out that both DeepSeek v4 Pro and v4 Flash massively beat Nemotron3 Ultra on agentic tasks, with v4 Flash using roughly a quarter of the active parameters.
SemiAnalysis argued that despite having brilliant talent, "committee-based model frontier development fundamentally does not work" for NVIDIA.
More from Companies & People
- Applied Compute Builds Internal AI Agent to Automate Post-Training Workflows — rhythmrg · 2026-08-13
- Microsoft's Homegrown AI Chip Effort Shows Signs of Life After Slow Start — pstAsiatech · 2026-08-13
- Researchers Call for Joining Independent AI Eval Orgs for High Impact — Afinetheorem · 2026-08-13
- e/acc Leader Beff Jezos: Never Bet Against Elon and xAI — beffjezos · 2026-08-13
- Paul Graham: Small, Fast Startups Are Safest Amid AI Disruption — Kyrannio · 2026-08-13
- CAICT: China's AI Industry Exceeds 1.2 Trillion Yuan in 2025 — nordicinst · 2026-08-13