https://adambutterworth.com/https://adambutterworth.com/about/https://adambutterworth.com/archives/https://adambutterworth.com/posts/https://adambutterworth.com/posts/attention-sinks/https://adambutterworth.com/posts/breaking-down-agent-evals/https://adambutterworth.com/posts/from-trpo-to-ppo/https://adambutterworth.com/posts/just-one-llm-bench/https://adambutterworth.com/posts/prompts-are-hyperparameters/https://adambutterworth.com/posts/setting-logits-to-negative-infinity/https://adambutterworth.com/posts/tau-bench-successors/https://adambutterworth.com/posts/tau-bench/https://adambutterworth.com/posts/why-recitation-works/https://adambutterworth.com/projects/https://adambutterworth.com/projects/competitor-research/https://adambutterworth.com/projects/nobet/https://adambutterworth.com/projects/studyzone/https://adambutterworth.com/search/https://adambutterworth.com/tags/https://adambutterworth.com/tags/agents/https://adambutterworth.com/tags/attention/https://adambutterworth.com/tags/benchmarks/https://adambutterworth.com/tags/context-engineering/https://adambutterworth.com/tags/decoding/https://adambutterworth.com/tags/ensembles/https://adambutterworth.com/tags/evals/https://adambutterworth.com/tags/inference/https://adambutterworth.com/tags/json/https://adambutterworth.com/tags/llm/https://adambutterworth.com/tags/mode-collapse/https://adambutterworth.com/tags/observability/https://adambutterworth.com/tags/policy-optimization/https://adambutterworth.com/tags/ppo/https://adambutterworth.com/tags/reinforcement-learning/https://adambutterworth.com/tags/structured-outputs/https://adambutterworth.com/tags/trpo/