Put the exact same code reviewer persona in the system prompt, the user message prefix, or both places at once. One configuration scores 88.5. Another scores 86.4. The instructions are...
Read more →In a previous experiment, we found that the “meticulous code reviewer” persona was the one persona that consistently improved AI code quality. A natural follow-up question: if one focused role...
Read more →“You are a senior software engineer with 15 years of experience.” If you’ve spent any time in prompt engineering circles, you’ve seen this advice. Persona prompting—giving the AI a role...
Read more →Sonnet loses 8 points with context noise. Opus gains 4. Your model choice determines whether bloated prompts hurt. I ran 1,080 benchmarks to answer a question that haunts every AI-assisted...
Read more →The conventional wisdom is simple: use the biggest model you can afford. Opus for quality, Haiku for speed, Sonnet when you want something in between. The assumption is that model...
Read more →“Think step by step.” It’s one of the most widely repeated pieces of prompt engineering advice. Add a chain-of-thought (CoT) instruction to your system prompt and the model reasons more...
Read more →There’s a running joke in the AI community about people who say “please” and “thank you” to their LLMs. The implication is that it’s sentimental, anthropomorphizing—treating a statistical model like...
Read more →I went to Game On Expo 2026 in Phoenix with one goal: see what a gaming convention’s RF environment actually looks like from the ground. One Samsung S21, WiGLE running...
Read more →The Next Knob to Test After I published “I Ran 540 Benchmarks to Prove CLAUDE.md Compression Works”, I got a recurring question: what about temperature? Fair question. If compressing your...
Read more →A few days ago I published “Compress Your CLAUDE.md”, arguing that you should strip human-readable formatting from your CLAUDE.md files to save 60-70% of context bloat. The logic seemed sound:...
Read more →