Art of Cyber AI — Research FeedField notes on reinforcement learning, LLM post-training, agent evaluation, and efficient inference.https://artofcyberai.com/2026-08-16T12:00:00-07:00Vikram Kharvihttps://artofcyberai.com/One machine. Two different jobs.https://trainrl.com/post-evals-rl-environments.html2026-08-15T12:00:00-07:002026-08-15T12:00:00-07:00Why post-training evaluation and reinforcement-learning environments need different abstractions, even when they run on the same machine.Inference begins with a service contracthttps://inference.artofcyberai.com/inference-service-contract.html2026-08-13T12:00:00-07:002026-08-16T12:00:00-07:00Define time to first token, inter-token latency, throughput, availability, workload shape, and cost before choosing serving hardware.Pre-training a language modelhttps://pre-trained.artofcyberai.com/tech-tokenization-model-design.html2026-08-13T12:00:00-07:002026-08-16T12:00:00-07:00A research runbook connecting corpus design, tokenization, Transformer training, distributed execution, checkpoints, and reproducible evidence.