Φ Institute for Physical AI @ BMI · Charlot Lab

Energy First Architecture

One energy that predicts, plans, remembers, verifies, and discovers.

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.

world model · predict by descending energy
planner · act by descending energy to a goal
memory · Hebbian fast-weights, written at inference
verifier · low energy = valid
discoverer · the law is the sparse energy that fits the data

The thesis

Invert every layer of the incumbent stack

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.

layerincumbentEFA inversion
unitpoint neuronsparse-positive, monosemantic
memoryfrozen weights + finite contextHebbian fast-weights, written at inference
legibilityblack boxmonosemantic / steerable by construction
world modelautoregressive next-tokenpredict consequences by energy descent
computedense matmul, one fixed passsparse, local, variable test-time thinking
substratecloud GPUon-device, energy-first, in a browser tab

What's proven — measured, and priced

The edge is narrow and real

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.

axisrepresentative result
the true Energy-Based Transformertrained 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 verificationenergy best-of-N → 100% selection; on real AR sequences too
zero-shot compositionconcept conjunctions never trained jointly, by energy summation
OOD generalizationgoal-agnostic energy 37→41% OOD where learned maps collapse
energy conservationHamiltonian NN, 5.5× lower energy drift
scientific discoveryLorenz · Burgers · Fisher–KPP · real lynx–hare data — laws recovered

The AI-scientist for physics

Energy-minimization is law discovery

Discover the equation

From noisy data, recover the governing ODE — a 2D oscillator, and the Lorenz chaotic system, exactly.

Discover the PDE

Burgers' (advection–diffusion) and Fisher–KPP (reaction–diffusion), recovered exactly from spatiotemporal fields.

Discover the invariant

A conservation law for the nonlinear pendulum — its energy — found from trajectories, correlation 0.99, without its form.

Discover from real data

Predator–prey Lotka–Volterra dynamics recovered from the real Hudson Bay lynx–hare records (1900–1920).

Relation to IPAI @ BMI and Physical AI

Why the Institute builds this

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

What is not claimed

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.

Energy First Architecture · Charlot Lab · Institute for Physical AI @ Bailey Military Institute.
Whitepaper · Validation ledger · Repository · Live research record · Ferric · Institute
Every figure here is a measured result, not an estimate.