There are times when I think that Large Language Models (LLM) are amazing and really do make my life a lot easier.
Then there are days, when they get really lazy and try sneaking in a short cut to avoid doing the hard work.
During a vibe coding session, I found a bug when I was testing the app and I asked the LLM to fix it.
After a short while, the LLM sent the following plan to me to ask for my approval:
The LLM wanted to fix the issue by just changing the syntax to avoid the error message.
Sometimes when the LLM is lazy, you just got to give it a kick! Ok, I'm not a violent person so I didn't smash my laptop, so this is what I did.
Then after more tokens were burnt, this response arrived:
*I really can’t stress enough how important it is to actually read through the plans and specifications an LLM produces. *
In vibe coding, even if we’re not writing the code ourselves, we still need to think critically and evaluate what’s being generated. If we just keep clicking “Ok” or “Proceed” without checking, there’s a good chance we’re letting bugs slip through that will surface at the worst possible moment.
Top comments (0)