Miguel Arroz2d ago
Jeff Johnson

Computer science experiment:

Put your LLM or agent through the same technical interview that you would require for prospective job applicants, and think about whether you would hire an applicant who gave the same responses as your LLM or agent.

If the answer is no, then why would you trust your LLM or agent to write code for you when you wouldn't trust a comparable human?

May 19 at 11:25amWeb
Show threadDrew McCormack2d ago
@lapcatsoftware Devil’s Advocate: Do you use a compiler, even though a compiler also wouldn’t pass the test? It’s a tool. It doesn’t replace, it enhances.
Show threadJeff Johnson2d ago
@drewmccormack I’m always puzzled by people who announce themselves as a devil’s advocate, as if they were performing some important charity work.
Show threadDrew McCormack1d ago
@lapcatsoftware Just trying to engage in discussion. But won't bother any more. I can honestly say I’ve never had a valuable interaction with you. Any sign of disagreement is just met with vitriol.
Show threadTaylor Hadden2d ago

@drewmccormack @lapcatsoftware What even is this response? People _do_ use coding agents as replacements. Blindly trusting the output is, and continues to prove to be, fraught.

The comparison to compilers is apples to oranges. They might both be fruit, but the similarities end there.