Hacker News Reader

Comment by jeroenvlek | Hacker News Reader

Had a similar experience. Asked Bard to generate a Firefox plugin and it responded that it was a text-based assistant and couldn't comply [1] whereas ChatGPT just did it. It didn't work, but it still generated it.

[1] https://twitter.com/jeroenvlek/status/1657195209686683648

View on HN

>It didn't work, but it still generated it.

So neither could generate the plugin, it's just that one of them lied about it.

No, I suspect it's more the case that ChatGPT could generate it anno 2021, but can't anymore due changes in Chromium (I switched to Edge ultimately) and the Gmail API and Google Cloud Credentials workflow that I tried to add to it later.

It was interesting to see that ChatGPT annotated its code with pitfalls to avoid, that it actually didn't avoid in the code.

>No, I suspect it's more the case that ChatGPT could generate it anno 2021

Why? I mean, why does it sound more plausible to you that it ever could, when we have countless examples of models lying (or rather hallucinating)?

Because it was trained on old documentation, but I see your point. My original point was also not necessarily that ChatGPT was correct, but that ChatGPT at least gave something to work with, whereas Bard simlly refused.

Just like a real human;)

The company I work for is building a browser extension that involves ChatGPT, and we can unfortunately confirm that it is, in fact, not capable of writing browser extension code.

Reply on news.ycombinator.com