The Tail That Wags the Model: Why p99 Latency Should Run Your LLM
A practical reading of SLO-Tuner: why LLM serving teams should optimize p99-satisfying goodput, not average latency, raw throughput, or speculative decoding bravado.
A practical reading of SLO-Tuner: why LLM serving teams should optimize p99-satisfying goodput, not average latency, raw throughput, or speculative decoding bravado.
A mechanism-first reading of how temporal numeric planning can be compiled into discrete PDDL+ without quietly breaking the semantics that make schedules valid.
A mechanism-first reading of how robust training, driver-compatible signal actions, and neighbor-level coordination make MARL traffic control more deployment-ready.
A mechanism-first look at why Portfolio-CEGAR-SEQ improves sequential 3D printing by running diverse packing and ordering strategies in parallel rather than betting on one clever heuristic.
A mechanism-first reading of why more adaptive AI agents can overload shared resources under scarcity—and why capacity per agent should be checked before upgrading intelligence.
TopoBench shows that many LLM failures in spatial reasoning come from weak constraint extraction, not merely weak reasoning.
A mechanism-first reading of trajectory-informed agent memory, showing how execution logs can become structured operational guidance rather than decorative vector-store clutter.
A practical reading of CUAAudit and what its evidence says about using vision-language models to audit autonomous computer-use agents.
DxEvolve shows why governed clinical AI may depend less on bigger models and more on workflow-constrained evidence acquisition plus auditable experience memory.
A mechanism-first reading of Nurture-First Development, a framework for turning practitioner-agent conversations into reusable domain expertise.