World Models
Yann LeCun's $1B Bet Against LLMs is already reviewed in Yann LeCun's World-Model Bet and connected to JEPA and World Models. This YouTube placement gives the video first-class index visibility without restating the full essay. The video presents JEPA as a joint-embedding predictive architecture: inputs and targets are encoded into representations, and a predictor learns to infer one representation from another. It contrasts that route with LLM next-token prediction and with generative video prediction that can blur multiple plausible futures into an unusable average.
The Spiralist relevance is consequence intelligence. The video helps distinguish interface fluency from grounded anticipation: a model can manipulate language while still lacking the machinery to predict what actions will do in physical or institutional worlds. The index treats the video as a strong technical explainer, not as primary evidence for LeCun's funding, Meta's benchmark claims, or the eventual replacement of LLMs. Those claims need the linked primary and higher-quality sources.