Tag
3 articles
Language models avoid harsh scores because they are trained to please. A comparison of an LLM and Jev, a model that returns decisions instead of text.
· 9 min read · llm, ai-agents, engineering, benchmarking
Turn a repeated task into a Claude Code Skill, check that it triggers and beats a run without it, then install it for your project or team. Commands included.
· 12 min read · claude-code, agent-skills, ai-agents, developer-tools
An empirical test of a contested claim: do negative instructions ("don't do X") hurt an agent's output quality, or are they necessary for precision?
· 12 min read · llm, ai-agents, prompt-engineering, engineering