Future foundation models will be built on egocentric, calibrated multiview data
ducha_aiki · x · 2026-09-09
Citing Jakob Engel's discussion of world models, duchaaiki argues that future foundation models will be built on and for egocentric, calibrated multiview, multimodal data — pointing to first-person sensor/video data as the training source for next-generation foundation models.
Related event: DeepMind Researcher on World Models and VLM Memory Reconstruction(4 posts)→
More from Models
- When will Clay Institute verify OpenAI's solution and pay the prize? — peterjliu · 2026-09-10
- GPT-6 shows 'incredible' physical-space instruction following on real robots, early tests show — k7agar · 2026-09-10
- Goodfire uses Ai2's open post-training stack to predict how training runs change model behavior — allen_ai · 2026-09-10
- Sebastian Raschka breaks down GPT-6 Astra rumors, looped transformers and hidden chains of thought — rhiever · 2026-09-10
- GPT-6 Astra's no-CoT time horizon estimated at 8 minutes to 1 hour — OwainEvans_UK · 2026-09-10
- LMArena to release human-ranked writing clarity results for Claude Fable 5 vs 5.1 — arena · 2026-09-10