The hierarchical architecture: the whole project aligned to the four patterns of the world model

LeCun has a slide with four lines: abandon X in favour of Y. We audited our code against each one. Abandon generative for joint-embedding. Abandon probabilistic for energy. Abandon contrastive for regularised. Abandon reinforcement for predictive control. For each line: which piece of our code implements it, which flag turns it on, and whether it is already measured. Where the project still does X and Y at the same time, it is written down with a number. And we kept the token head on purpose, for three reasons LeCun's talk does not cover. One of them: removing the head today would mean removing the measuring instrument.

The three points

1. Energy: candidates are reordered by plausibility plus distance to the heading. Implemented in the energy, gradient and mpc modes.
2. Predictive control: beam search in latent space is MPC in miniature. And when the plan does not give the predicted result, the map is corrected, not the generator. Error ratio fell from 1.303 to 0.713 and generalised.
3. Why the token head stays: it produces text, gives the guardrail that keeps the planner inside the distribution, and is the only ruler comparable with everything that exists.

From the chapter, the beginning

The whole project, piece by piece, aligned with the four patterns of the world model: joint-embedding, energy, regularization and predictive control — and what is still hybrid, stated with numbers.

The slide "Recommendations to my fellow AI scientists" has four lines, each an abandon X in favor of Y. This chapter is an audit: for each line, which piece of our code implements it, which flag turns it on, and whether it has been measured yet. Where the project still does X and Y at the same time, that is written down — because the house rule is keep and improve, without regression, not swapping what works for a promise.

The talk argues that predicting symbols is not enough to build a world model. Even so, our token decoder does three things the world model alone does not: it produces text (the latent is not readable), it provides the guardrail −log p that keeps the planner from leaving the distribution, and it is the only ruler comparable with everything that exists (bits/character). Removing the token head today would be removing the measuring instrument. And there is a stronger reason, which the interview gives (chapter 10, §3): the intractability that motivates JEPA holds for continuous signals; in text, discrete and finite, the softmax over the next token is the distribution over the futures. The plan is to invert its role — from decision to energy term — and that is what the energy and mpc modes already do.