Capabilities that have crossed the line. Oldest flip first — a chronology of the frontier.
The snake-eating-its-tail phase of ML — most foundation models now train partly on synthetic data generated by their predecessors. --- AI can generate plausib...
Khan Academy's AI tutoring layer launched this in 2024–2025 — auto-generated explainer videos, adaptive to the student's prior knowledge. --- Currently, AI ca...
Mostly working at the indie scale — Godot/Unity scaffolds, simple gameplay loops, art assets. Years away from AAA, but fully real for game jams. --- Current A...
Models with retrieval over the literature now write surveys with proper citations, identifying gaps and trajectories that human reviewers verify. --- AI can g...
Score-following AI synced to picture, themes for characters, tonal arcs. Indie filmmakers swap in AI scores during edits and often keep them. --- AI systems h...
Auto-cuts, B-roll selection, music sync. Most YouTube essay channels run AI editing pipelines now. --- Current AI systems can perform various tasks related to...
Capability and societal harm scaled together. Detection is now a parallel arms race. --- AI can generate realistic deepfake videos of public figures speaking,...
DeepMind's AlphaProof + AlphaGeometry 2 reached silver-medal level at IMO 2024 and approached gold by 2025 in geometry and number theory. --- AI systems have...
Precision-medicine assistants used at major academic medical centers. Final decisions remain with clinicians; suggestions are good enough that ignoring them cos...
45 minutes of cohesive music with thematic continuity. Spotify's catalog grew by ~20% from AI-generated tracks in 2025 alone. --- AI systems have made signifi...
DeepMind's RT-2 and successors showed end-to-end vision-language-action models could follow a recipe with mistakes mostly within human range. --- AI can run a...
Multi-agent autonomy at scale — research and military deployments demonstrated this through 2025–2026 in increasingly complex environments. --- Researchers ha...
Agentic systems run multi-step web tasks, file ops, calls to other agents. Not yet reliable enough for all jobs, but solidly working for many. --- Current res...
Add a statement to the atlas. We review weekly.
Site
stuffaicantdo.com © 2026 · made in NL by Arcadist
Pick a side, give it a one-line title, optional explanation, and a category. We review submissions weekly.