Post #4447812

If you must use LLMs (and you really mustn’t), the human should always be the driver of the task. ❌ “Summarize this document.” ✅ “Where in this document does it describe xyz?” ❌ “Write this code for me.” ✅ “Tell me why my code isn’t producing the result I expect.” At most, LLM output should guide human decisions and output, not produce the output. “I’ll check its output” is a fail-dangerous setup. “I’ll do it myself with hints from the LLM” is fail-safer.