Taalas 14,000 tokens/sec demo comes from The Peter McCormack Show
iamfakhrealam · x · 2026-08-31
Follow-up: the widely shared Taalas inference demo (100 tokens/sec for ChatGPT vs 14,000 tokens/sec for Taalas) originates from the YouTube channel The Peter McCormack Show, where Emad Mostaque demonstrated the model baked directly into silicon live on air.
Related event: Taalas demos 14,000 tokens/sec inference, ~100x faster than ChatGPT(3 posts)→
More from Infra
- Europe invests €387.8M in LUMI-AI supercomputer with 10x AI capacity — wkmyrhang · 2026-09-01
- SK hynix reportedly considers Intel Foundry for HBM4E base dies — AccBalanced · 2026-09-01
- Guardian proposes off-grid, self-powered datacenters to cut emissions — nordicinst · 2026-09-01
- Compute Wants to Leave Earth: A Manifesto for Orbital Infrastructure — McDonaghMatthew · 2026-09-01
- Optimizing Qwen 3.8 Flash Next: Improving speeds on 64GB VRAM setup — Jorlen · 2026-09-01
- How much power does the AI buildout actually take? — TheZachMueller · 2026-08-31