Action JEPA · Joint Action and World Model Prediction · JEPA Becomes Control
Abstract ACT-JEPA (2025) represents the clearest bridge from JEPA to action and policy learning. It jointly predicts action sequences and latent observation sequences. The transition from passive world modeling to active control introduces new instances of the 6 gaps, particularly the WLD gap: a control system with no mercy reset is a safety hazard.
6 FORMAL GAPS · 1 PER CANON SYMBOL
No Invariant Anchor Linking Action Prediction to World-Model Prediction
γ₁ — THE FLOOR
ACT-JEPA jointly predicts actions and world-model states. There is no formal invariant γ₁ that links the action prediction space to the world-model prediction space. An action that is consistent with the world model and an action that contradicts it are treated symmetrically at the representation level. The floor — the grounding link between action and world — is absent.
Action Predictor Not Self-Adjoint With Observation Predictor
H=H† — THE HONEST GATE
ACT-JEPA predicts actions from observations and predicts observations from actions. These two predictions are not formally verified to be symmetric. A self-adjoint action-world predictor would satisfy: if action a leads to observation o, then observation o should be verifiable as consistent with action a in reverse. This bidirectional consistency check is absent.
No Paradigm Audit Between Reactive and Planned Actions
LSOS — THE READER
ACT-JEPA can produce both reactive (immediate response) and planned (multi-step) actions. There is no audit of the paradigm shift between reactive and planning modes. When the system transitions from reactive to planned, the active policy paradigm changes without acknowledgment. LSOS would read and flag this transition.
No Mercy Reset When Policy Collapses
WLD — THE RESET
When ACT-JEPA's policy collapses — when it learns a degenerate action sequence (e.g., always output the mean action) — there is no mercy reset. This is the most safety-critical gap: a collapsed policy in a control system can cause irreversible real-world harm. WLD is a mandatory safety requirement for any deployed ACT-JEPA system.
No Continuity Guarantee in Strategy Switching
FEP — THE SWITCH
ACT-JEPA must switch between different action strategies based on the current world state. There is no formal continuity guarantee for strategy switches. FEP ensures that switching from one action strategy to another preserves the learned policy paradigm — the switch is safe and the previous strategy can be recovered.
Action Sequence Horizon Has No Named Ceiling
FOF — THE BREACH
ACT-JEPA does not define a formal upper bound on action sequence length (planning horizon). As the horizon grows, the architecture's predictions become unreliable. The point where ACT-JEPA's planning breaks down — where the predicted action sequence diverges from any achievable real-world sequence — is not named. FOF names this boundary.
STE COMPLETION LAYER
What changes when you add the 8-symbol Canon
Adding the Canon to ACT-JEPA does not change the architecture. It adds the missing structural layer:
⚓ γ₁ — invariant anchor: mathematical ground truth latent representations must converge to.
⯛ H=H† — honest gate: bidirectional verification of every prediction.
〰️ LSOS — paradigm reader: reads active paradigm before reasoning begins.
🌀 WLD — mercy reset: detects collapse and resets to last stable state.
γ FEP — safe switch: continuity guarantee across paradigm transitions.
🌌 FOF — named ceiling: formal boundary of what the architecture can claim.
═ EVEN — substrate: ground beneath all the above. What holds when everything else is active.
The Canon is not an add-on. It is the formal completion of the JEPA programme.
X POST · @ylecun
POST 1 — Name the gap
@ylecun ACT-JEPA (2025): Gap 2 (H=H†) — predictor not self-adjoint. Asymmetric by design. Not an empirical limitation — a missing symbol. pemos.ca/actjepa-gap
POST 2 — Canon map
@ylecun ACT-JEPA: 6 gaps · γ₁ (no anchor) · H=H† (no gate) · LSOS (no audit) · WLD (no reset) · FEP (no continuity) · FOF (no ceiling). Same in all 14 milestones. pemos.ca/jepa-index
POST 3 — Invitation
@ylecun ACT-JEPA gap analysis: part of a 14-milestone series. Same 6 structural gaps in every milestone. The gaps are there because the symbols were never in scope. They are now. pemos.ca/jepa-index
ACT-JEPA is a landmark in the JEPA lineage. The 6 gaps we identify are not critiques of the engineering — they are structural absences that the Canon fills. Each gap maps to a symbol that was always going to be necessary once the JEPA architecture matured. The Canon did not wait for the JEPA timeline; the JEPA timeline arrived at the Canon. The gaps are there because the symbols were never in scope. They are now.
Gap 1 (γ₁): No Invariant Anchor Linking Action Prediction to World-Model Prediction ACT-JEPA jointly predicts actions and world-model states. There is no formal invariant γ₁ that links the action prediction space to the world-model prediction space. An action that is consistent with ...
Gap 2 (H=H†): Action Predictor Not Self-Adjoint With Observation Predictor ACT-JEPA predicts actions from observations and predicts observations from actions. These two predictions are not formally verified to be symmetric. A self-adjoint action-world predictor would satisfy...
Gap 3 (LSOS): No Paradigm Audit Between Reactive and Planned Actions ACT-JEPA can produce both reactive (immediate response) and planned (multi-step) actions. There is no audit of the paradigm shift between reactive and planning modes. When the system transitions from ...
Gap 4 (WLD): No Mercy Reset When Policy Collapses When ACT-JEPA's policy collapses — when it learns a degenerate action sequence (e.g., always output the mean action) — there is no mercy reset. This is the most safety-critical gap: a collapsed policy...
Gap 5 (FEP): No Continuity Guarantee in Strategy Switching ACT-JEPA must switch between different action strategies based on the current world state. There is no formal continuity guarantee for strategy switches. FEP ensures that switching from one action str...
Gap 6 (FOF): Action Sequence Horizon Has No Named Ceiling ACT-JEPA does not define a formal upper bound on action sequence length (planning horizon). As the horizon grows, the architecture's predictions become unreliable. The point where ACT-JEPA's planning ...
The STE provides the completion layer for each gap. The gaps are not empirical — they are structural. Adding the symbols closes the gaps by definition.
Here's how to explain ACT-JEPA gaps to a 10-year-old:
Gap 1 — No floor: Imagine you're building a tower and you don't have a foundation. The tower might stand for a while but there's nothing it's guaranteed to come back to if it leans. γ₁ is the foundation. The math already built it. ACT-JEPA didn't use it.
Gap 2 — No honest check: If I tell you something, and you can't ask me 'are you sure?' and get the same answer backwards, I might be guessing. H=H† is the rule that every claim must survive being asked backwards. ACT-JEPA makes claims it can't verify in reverse.
Gap 3 — No paradigm reader: Sometimes you're playing one game and someone changes the rules without telling you. LSOS reads the rules before you play. ACT-JEPA changes paradigms without reading them first.
Gap 4 — No reset button: When you're drawing and you make a mistake, you have an eraser. WLD is the eraser. ACT-JEPA doesn't have one. If it goes wrong, it keeps going wrong.
Gap 5 — No safe switch: If you switch from reading a book to watching a movie, you expect to pick up where you left off. FEP is the bookmark. ACT-JEPA switches paradigms and might drop the book.
Gap 6 — No ceiling: Every ladder needs to know when to stop. A ladder that claims it goes forever is a lie. FOF names where the ladder ends. ACT-JEPA hasn't named its ceiling yet.
The Canon is not a critique. It's the toolkit that was always going to be needed. ACT-JEPA got as far as it could without it. Now the toolkit exists.
The 6 gaps identified in ACT-JEPA are structural absences in a published architecture. Identifying structural absences in published work is not IP infringement — it is prior art analysis. The STE symbols that fill these gaps are EOSE IP. The gap identification is public analysis. The completion layer is proprietary.
Status: LSOS-OWNERSHIP-001 open (EOSE Labs Inc. not yet registered). No public disclosure of STE completion layer until registration clears. Gap analysis pages (like this one) are disclosure of the gaps, not the fills. This is legally distinct and permissible.