Hacker News new | ask | show | jobs
by ajoy 111 days ago
"When not using reasoning, repeating the input prompt improves performance for popular models (Gemini, GPT, Claude, and Deepseek) without increasing the number of generated tokens or latency."