Editorial Search Archive

Cover NVIDIA AIPerf launch: multiprocess inference benchmarking with TTFT, ITL tails, and realistic traffic that ends vanity throughput figures for good.

Build guarded text-to-SQL agents with read-only roles, cost ceilings, and propose-verify-repair loops that hold 98.1% valid queries with zero escapes.

Route test-time compute by difficulty: match best-of-16 accuracy within 0.01 points at 58.9% fewer tokens with calibrated early stopping and halved latency.

Cover GPT-5.4 Pro FrontierScience lead at 36.7%: rubric-graded PhD research subtasks split from saturating Olympiad theory, plus the builder playbook.

Cover GLiFormer 575M encoder release: 91.10 F1 extraction past GPT-5-mini with zero generated tokens, plus the LLM-to-encoder migration playbook.

Run TDD agents with test-impact maps: human repro tests unlock 94.3% resolution while bare TDD prompting raises regressions 63% — maps over mantras.

Deploy a PagerDuty on-call MCP server with read-open triage, urgency-gated writes, and diagnosis notes that land before responders wake with paging untouched.

Ship cron agents that survive the night: overlap policies, atomic outbox idempotency, and heartbeat absence alerts with zero duplicate side effects.

Cover the Vercel record: open weights take 78.4% of gateway tokens as Moonshot, DeepSeek, Z.ai outspend OpenAI — plus the mix-audit playbook for builders.

Settle the stuff-vs-retrieve debate with measurements: 8K effective windows lose to graded RAG, and a Self-Route hybrid holds quality at 34% of cost.

Build a FastMCP disclosure server that loads 60 tools on semantic triggers, cutting init context from 48,000 to 2,000 tokens with cacheable manifests.

Ship adaptive compaction for coding agents: fire at phase transitions, preserve five-field state, and hold 97.8% next-action accuracy at 0.3x cost.