Energy First Architecture
The dominant AI stack — point neuron, dense matmul, frozen weights, a black box in a cloud datacenter — was built for machines that talk. Physical AI is for machines that act, and acting is an energy problem. EFA takes energy as the native object: a single learned scalar energy, over a sparse-positive latent, does what the incumbent stack assembles from separate subsystems.
The thesis
Each inversion has a research lineage; the claim is unification — not six modules, but three mechanisms over one representation, because a sparse-positive activation space is at once the readable feature space, the associative-memory space, and the world-model prediction space.
| layer | incumbent | EFA inversion |
|---|---|---|
| unit | point neuron | sparse-positive, monosemantic |
| memory | frozen weights + finite context | Hebbian fast-weights, written at inference |
| legibility | black box | monosemantic / steerable by construction |
| world model | autoregressive next-token | predict consequences by energy descent |
| compute | dense matmul, one fixed pass | sparse, local, variable test-time thinking |
| substrate | cloud GPU | on-device, energy-first, in a browser tab |
What's proven — measured, and priced
Built the way it should be judged: prove each mechanism, compose, scale, push to hard tasks and real data — pricing every limit. EFA is not claimed to beat frontier systems at scale. Its demonstrated edge is on the axes where a native energy is the right representation.
| axis | representative result |
|---|---|
| the true Energy-Based Transformer | trained through an unrolled energy descent (2nd-order autograd): 100% where feedforward is 0%; thinks (K=1→6: 22→100%) |
| test-time thinking (acting) | MPPI latent planning 39% → 69% reach, same value net |
| native verification | energy best-of-N → 100% selection; on real AR sequences too |
| zero-shot composition | concept conjunctions never trained jointly, by energy summation |
| OOD generalization | goal-agnostic energy 37→41% OOD where learned maps collapse |
| energy conservation | Hamiltonian NN, 5.5× lower energy drift |
| scientific discovery | Lorenz · Burgers · Fisher–KPP · real lynx–hare data — laws recovered |
The AI-scientist for physics
From noisy data, recover the governing ODE — a 2D oscillator, and the Lorenz chaotic system, exactly.
Burgers' (advection–diffusion) and Fisher–KPP (reaction–diffusion), recovered exactly from spatiotemporal fields.
A conservation law for the nonlinear pendulum — its energy — found from trajectories, correlation 0.99, without its form.
Predator–prey Lotka–Volterra dynamics recovered from the real Hudson Bay lynx–hare records (1900–1920).
Relation to IPAI @ BMI and Physical AI
On-device, everywhere. Physical AI must act at the edge, at low energy. EFA runs in a browser tab on WebGPU via the Institute's pure-Rust Ferric compute layer — the substrate inversion, made real.
Energy is the native language of the physical world. Robotics, control, and the physical sciences are already energy-based. EFA doesn't translate physics into a loss; it represents it as an energy — which is why the discovery suite works cleanly, the prior native rather than bolted on.
Legible, and teachable. Every mechanism is a small, driveable, on-device artifact. EFA is a research bet and a curriculum surface at once — the Institute teaches the ideas by letting students run them.
Honest limits
Scale. Everything is nano-to-small. EFA is not claimed to beat frontier transformers at in-distribution capability or frontier scale — the edge is narrow, not raw capability.
Sampling. Energy descent (generation) is step-size sensitive; the robust route is verification (best-of-N) or Metropolis correction — priced, not hidden.
A settled negative. An energy-conserving surrogate at field scale did not beat a naive force net — and it still lost after Ferric gained second-order autograd and the test was redone with exact gradients. So the negative is structural, not a tooling gap. (That same autograd unlocked the true EBT above.)
AI-for-math. The energy verifier is mechanism-sound but gated on a Lean-task-trained encoder; general embeddings sit at chance, because tactic↔goal compatibility is a formal, not surface-semantic, property.