Ask HN: Have people started to see cracks in the ChatGPT answers with time
3 points
20 hours ago
| 3 comments
| HN
Context: I have a co-worker who is used to execute the suggested scripts from Chat-gpt with high confidence. Lately one of the scripts to use Github API was not working and he asked me to debug. Turns out, Github had changed/updated the attribute names in the output and they no longer matched with the Chat-Gpt provided script. Hard part, this individual is really good engineer, however he/she was low on confidence about how to debug the Chat-gpt provided script
almosthere
20 hours ago
[-]
Don't use chatgpt for that kind of work, use agents. It will read the error message and adapt, perhaps even go as far as open the release notes for the tools you're using and fix the problems it encounters.
reply
bjourne
16 hours ago
[-]
IMnsHO, nope! People trust the bot way more than their own experience.
reply
speedgoose
20 hours ago
[-]
Skill issue.

First, switch to Codex. Then, consider using the context7 mcp server to let agents automatically fetch the latest documentation.

Debugging is not so different than debugging code written by someone else. It’s a skill you can practice.

By the way, agents are also good at debugging.

reply
techblueberry
19 hours ago
[-]
This is a great advertisement:

Agents: harder, slower and more expensive than just doing it yourself.

reply
pavelai
4 hours ago
[-]
It's a very overgeneralized statement. Agents could do a lot of work, which most of people wouldn't want to do. and agents can do this on a pretty good level for now. Probably we would enhance their ability to do this in near future. People should stop write code and start to do real life things.
reply
techblueberry
3 hours ago
[-]
What you are failing to understand, is that this is literally the worst these models will ever be. Uber wasn’t profitable for like the first 20 years of their existence, and now look at them. People said the same things about the internet.
reply
speedgoose
9 hours ago
[-]
It depends on the task.
reply