fal engineering head: we'll never pre-train, inference compute is the real moat
jfischoff · x · 2026-09-17
fal's head of engineering explains why the generative media cloud will never pre-train a model: the gaps today are in real-time video generation — which fal claims to have cracked with H3 Max — not pre-training, and "nobody in the world has enough inference compute," so fal is aggressively building and procuring compute. The reposter adds that capabilities really emerge in post-training, where they doubled down last year and saw it pay off.
More from Companies & People
- Reducto opens NYC office as monthly document processing tops 1 billion pages — JenniferHli · 2026-09-17
- DeepMind co-founder Shane Legg launches DeepMind Institute to study AGI implications — AllanDafoe · 2026-09-17
- Open weights are not open source: why AI's favorite label is under dispute — StanfordHAI · 2026-09-17
- PyTorch Conference NA lineup spotlights torch.compile and custom kernel breakthroughs — PyTorch · 2026-09-17
- New Class of AI 'Judgment Models' Like Jev Could Reshape Business Automation — The AI Daily Brief · 2026-09-17
- Princeton's Tom Silver group and Basis are hiring a robotics postdoc — tomssilver · 2026-09-17