AI Explained
Plain explanations of trending AI concepts, with live visualizations.
LLM-as-judge bias appears as activation geometry — Steerable bias directions — What does it mean?
LLM-as-judge bias is a steerable direction in the judge's hidden state — a study that reads it, predicts it, and steers it back to baseline.
OpenAI trains GPT-Red to harden agents against prompt injection — Self-play red-teaming — What does it mean?
OpenAI trained an attacker model to invent prompt injections, then fed everything it found back into the defender's training — a red-teamer that never ships.
vLLM 0.25.1 stops a fused kernel from corrupting NVFP4 models — Mixed-dtype quant-fusion guard — What does it mean?
vLLM 0.25.1 adds a dtype check so a fused allreduce+RMSNorm+quant kernel stops silently corrupting mixed-precision NVFP4 models.
E3 cuts coding-agent scope before expanding context — Estimate-Execute-Expand — What does it mean?
E3 is a coding-agent strategy: start from the smallest scope, expand only when a check fails, cutting cost ~85% at the same success.
AVQ-Attention refines codewords where attention mass concentrates — Adaptive vector-quantized attention — What does it mean?
AVQ-Attention represents keys as a few codewords, then adds detail only where attention concentrates — turning O(N²) attention into O(MN).
KronQ adds gradient covariance to LLM quantization — Kronecker-factored Hessian — What does it mean?
KronQ scores the rounding error by its output impact, not just its input size — so 2-bit LLaMA-3-70B holds instead of collapsing.
Interaction scaling grounds agent feedback loops — Instrument-grounded feedback loops — What does it mean?
An agent keeps improving when it acts, measures the result with a tool, and revises — not by thinking longer or trusting a review by eye.
HCRMap places hot MoE experts across chiplet memory tiers — Hotness-aware MoE expert replica placement — What does it mean?
HCRMap ranks each MoE expert by how hot it runs, then promotes or evicts its copies across memory tiers, cutting latency up to 46.7%.
GATS plans agent tasks with zero LLM calls during the search — World-model tree search — What does it mean?
GATS plans an agent's actions by searching a learned world model — the search makes zero LLM calls, and it solved every task in the paper's test.
FastTPS fuses token-phase LLM inference on AI accelerators — Reloading-free KV-cache concatenation — What does it mean?
FastTPS runs LLM decode without recopying the KV cache to append: tiled attention then hits 93% peak memory bandwidth, 6× faster on an NPU.
ARMT extends LLM context with recurrent associative memory — Associative recurrent memory — What does it mean?
ARMT extends an LLM's context with a fixed-size recurrent associative memory instead of a growing KV cache — flat memory as text grows, ~30% fewer FLOPs.
Self-guided TTT adapts long-context LLMs on selected spans — Evidence-span test-time training — What does it mean?
Before answering, the model highlights the evidence spans that matter and briefly trains on only those — a per-question tune-up that recovers long-context accuracy.
