Artificial Atlas

Who is betting on what in AI, and against whom.

The four fault linesEvery camp is its answer to these four questions.

Rows are camps, columns are the questions. Product categories (chatbots, robots, video) cut across these; the questions are what the camps actually disagree about.

SignalWhat does the model learn from?WhenWhen does learning happen?RepresentationWhat is the internal representation?Acts inWhere does it act?
Scale next-token prediction

Text, then verifiable rewards and human preference

Internet-scale textVerifiable outcome rewardsHuman preference

Offline pretrain, then freeze; memory is a retrieval layer

Pretrain once, then freeze

Tokens over a transformer

Tokens over a transformer

Software: chat, code, browsers, APIs

Software
World models

Self-supervised prediction of video and latent state

Video and sensor streams

Offline pretraining, but built to plan at inference by rolling out imagined futures

Pretrain once, then freezePlan at inference by rolling out imagined futures

Latent embeddings (JEPA) or generated pixels (video models). These sub-camps disagree.

Latent embeddings (JEPA)Generated pixels (video models)

Simulated and physical environments; robotics and driving downstream

Simulated environmentsThe physical world
Physics AI and AI for science

Simulation data plus the governing equations themselves, which act as a checkable reward

Simulation dataGoverning equations as a checkable reward

Offline pretraining, then optimization loops at inference: simulate, take a gradient, redesign

Pretrain once, then freezeOptimisation loops at inference

Continuous fields in 3D plus time; neural operators, resolution-invariant

Continuous fields in 3D plus time

Design and discovery loops: materials, fusion, weather, devices, drugs

Design and discovery loops
Learn from experience

Scalar reward from interaction, plus self-generated subgoals (options)

Scalar reward from interaction

Continually, at runtime, forever. No train/deploy boundary

Continually at runtime, no train/deploy boundary

Whatever the agent builds online: learned features and options, no replay buffer

Features and options built online

Any environment, ultimately embodied. Target: 20 watts

The physical worldSimulated environments
Physical AI and robotics

Teleop demonstrations, human video, sim-to-real, then on-robot RL

Teleoperated demonstrations and human videoSimulation dataScalar reward from interaction

Offline, with on-robot RL fine-tuning starting to matter

Offline, then on-robot fine-tuning

Mostly camp 1: a vision-language model with a diffusion or flow action head

Vision-language model with an action head

Actuators in the real world

The physical world
Program search and other substrates

Task success on novel abstraction (ARC), free-energy minimization, symbolic consistency

Task success on novel problemsFree-energy minimization

Often at test time: search and adapt per task rather than pretrain once

Search and adapt per task at test time

Explicit programs, symbols or probabilistic models; deep nets only as a guide for search

Explicit programs and symbolsProbabilistic models

Mostly abstract problems today; some robotics via active inference

Abstract problemsThe physical world

Allowed values

Signal
  • Internet-scale text text
  • Human preference human-preference
  • Verifiable outcome rewards verifiable-reward
  • Video and sensor streams video
  • Simulation data simulation
  • Governing equations as a checkable reward physical-law
  • Scalar reward from interaction scalar-reward
  • Teleoperated demonstrations and human video demonstrations
  • Task success on novel problems task-success
  • Free-energy minimization free-energy
When
  • Pretrain once, then freeze offline
  • Plan at inference by rolling out imagined futures inference-planning
  • Optimisation loops at inference inference-optimisation
  • Offline, then on-robot fine-tuning on-robot-finetune
  • Continually at runtime, no train/deploy boundary continual
  • Search and adapt per task at test time test-time
Representation
  • Tokens over a transformer tokens
  • Latent embeddings (JEPA) latent
  • Generated pixels (video models) pixels
  • Continuous fields in 3D plus time fields
  • Features and options built online online-features
  • Vision-language model with an action head vla
  • Explicit programs and symbols programs
  • Probabilistic models probabilistic
Acts in
  • Software software
  • Simulated environments simulation
  • The physical world physical
  • Design and discovery loops design-loops
  • Abstract problems abstract