Hacker News Reader

Comment by TrainedMonkey | Hacker News Reader

[flagged]

Why would ChatGPT have any idea about this at all? Its dataset wasn't made up of estimations for energy funds, and it makes absolutely no sense to ask it about that? I'm not sure why these figures or even their ratio would be trustworthy.

ChatGPT didn't use it's dataset for the numbers sourced in the linked chat. That said you do have to still go validate the sources it finds and see they say it says what they say it says, which I don't think is true for the second claim.

It’s terrifying that people are using chatgpt for questions like this. Isn’t it obvious what it is and isn’t good for?

It's more terrifying that people have claims about things they didn't even check. Why exactly is it bad? Do you really think that a human quickly skimming few web pages is better than a bot quickly skimming few web pages? Or did you think it's pulling the data from its trained weights?

People slam this for being ChatGPT but I wonder how often the references it finds on the web are more relevant than ones in this [1] style of HN [2] comment that often get to the top of this type of thread yet lack any deep analysis either. Slamming it for not validating the sources support the claim at least makes sense but isn't really specific to the ChatGPT fixation.

Perhaps somewhat telling on that is the comments are focusing on how the source was found rather than debating or correcting the figures themselves. The first figure sourced seems both accurately sourced and accurately estimated. The second figure I'm having trouble finding in the linked source and the figure seems to be more based on renewable energy investment not renewable energy research. I'm also having a hard time finding any good source for current renewable energy research investment to replace that with though so I can't say much in regards to how the final conclusion fairs.

The problem is that even when incorrect, typically someone intended the correct information to on a website.

Where LLMs are just an inference of related words.

So by your scenario, the LLM also picked up the incorrect answer on a website and included it into its inference.

He isn’t downvoted for being lazy, he’s downvoted for using a language model as a fact model. Even if sometimes it does work.

The problem has nothing to do with how you're finding your sources and everything to do with the lack of bothering to verify and understand them properly. What does it matter if you got your source list from tea leaves so long as you validate the information in the sources?

Nothing replaces actually going out and verifying anything yourself (yet) but that doesn't make tools which do a first past somehow to blame for someone's laziness while tools that put no effort in at all are supposed to be just fine.

That is an apples to oranges comparison. You are comparing r&d on an emerging technology to deployment of a mature technology. Its like spending on developing a new plane to purchasing an existing one.

We need to rebrand chatgpt and this generation of AI as "the Internet (tm)."

"Asking the Internet and it says..." Well, yeah, the Internet is full of crazy ideas and people untethered from reality and scientific inquiry. But enough people have put samples of things we do know, so sometimes The Internet gets things right.

ChatGPT fails miserably at questions like this. It cherrypicks various data points and arbitrarily runs sums.on them, convoluting nuance and meaning, conflating disparate or overlapping measures, and presents them as fact.

Always ask it to quote all its sources, and check them.

You've been heavily downvoted for good reason.

Reply on news.ycombinator.com