AI Explained
Plain explanations of trending AI concepts, with live visualizations.
BlockSearch traces million-token retrieval collapse — Attention dilution — What does it mean?
BlockSearch traces long-context retrieval collapse to attention dilution — the softmax denominator drowning the gold document — and fixes it with a length-aware softmax.
AgenticSTS tests long-horizon agent memory — Bounded-memory contract via typed retrieval — What does it mean?
AgenticSTS reframes agent memory as a contract — each decision sees only what typed retrieval hands it, so the prompt stays bounded and each memory layer is testable.
TRIAGE cuts agent turns up to 14.8% — Role-typed credit assignment — What does it mean?
TRIAGE grades each agent action by its role — progress, probe, stall, or blunder — not just the final win, so RL wastes fewer turns.
SkillHone evolves agent skills across sessions — Persistent decision-history memory — What does it mean?
SkillHone improves an agent's skills across sessions from an external decision history it reads — no retraining, no weight changes.
QVal: training-free testbed finds prompting beats dense agent supervision — Q-aligned dense supervision — What does it mean?
QVal tests, with no training, whether an agent's per-step supervision ranks actions like a reference policy's Q-values — and finds prompting beats 21 methods.
Orca learns a unified world model over video and events — Next-State-Prediction — What does it mean?
Orca trains one shared world model with a single objective — predict the next state — then reads it out as text, image, or action.
CausalMix picks LLM pretraining data mixtures via causal inference — CATE-based data mixture selection — What does it mean?
CausalMix picks an LLM's pretraining data mix by causal inference: estimate each mix's causal effect from 512 tiny runs, then scale to 7B.
ELDR routes MoE decode by expert locality, cutting TPOT up to 13.9% — Expert-locality-aware decode routing — What does it mean?
ELDR predicts a request's experts and routes decode to the worker whose zone matches, so fewer experts reload — cutting TPOT 5.9–13.9%.
DOPD dodges the 'privilege illusion' — Dual on-policy distillation — What does it mean?
DOPD routes each token's distillation signal by advantage, so a small model learns real skill instead of an answer key it can never hold.
Dockerless verifies coding-agent patches without containers — Execution-free patch verification — What does it mean?
Dockerless judges whether a coding agent's patch is correct by exploring the repo and reasoning about it, no Docker tests — 62.0% on SWE-bench Verified.
BlockPilot gives diffusion speculative decoding 4.2× — Instance-adaptive draft block sizing — What does it mean?
BlockPilot learns a per-input policy that predicts the best draft block size for diffusion speculative decoding, hitting 4.20× on a 4B model.
OSWorld2.0 benchmark: best computer-use agent finishes just 20.6% of tasks — Long-horizon computer-use failure modes — What does it mean?
OSWorld2.0 runs computer-use agents through 108 long real-world tasks. The best finishes just 20.6%, undone by four failure modes.
