[spideymans]

My ChatGPT use is down. After the novelty wore off, it quickly became apparent that ChatGPT was exceedingly willing to outright lie to you. Which makes it not useful, unless you’re already a subject matter expert who can spot its lies. But if you’re a subject matter expert, you probably don’t need ChatGPT to help you in the first place; I can find answers on Google faster than ChatGPT can generate them.

I suppose it’s also useful for generating things like letters (things where exact truthiness doesn’t really matter). But even for that use case, I find ChatGPT creates overly verbose corporate gobbledygook whenever I ask it to generate text. So I just end up writing the text myself.

[rsynnott]

> I find ChatGPT creates overly verbose corporate gobbledygook whenever I ask it to generate text

I feel like this doesn't get enough attention; its writing style is _incredibly_ grating. It's not just corporate gobbledygook; it's a parody of corporate gobbledygook. Just awful.

[spideymans]

If you ask ChatGPT an exceedingly trivial question, it’ll typically spend the next 60 seconds spewing out five paragraphs of corporate gobbledygook. And of course, because ChatGPT will lie to you, I often end up back on Google anyways to validate it’s claims.

Meanwhile, I could’ve found conclusive and correct answers directly from Google in about 10 seconds (I’m a fast Googler).

There are exceedingly few situations where I find ChatGPT is worth the effort. At least for factual Q&A-style queries like this.

[ragequitta]

I think the pull for most of us who use chatgpt is that google lies far, far more often than chatgpt ever will. Or is just otherwise inconclusive / does not give the relevant information you're looking for. The amount of SEO clickbait or quora/stack overflow answers that are either just incorrect or highly opinionated makes google very difficult to use for many things. As someone new to/learning Fedora it gives me the right answer 95% of the time, google gives me the right answer in the top 5 links far less.

[AnimalMuppet]

Very much this. Back in the day, the Cluetrain Manifesto said that corporate writing sounded "literally inhuman" - like it didn't come from a human being. You can't hear the person in it.

ChatGPT reads similarly. There's no personal voice. It's like food with no seasoning; it's just... blah.

[zikzak]

I write fiction for fun (just for me, don't ask) and asked it to generate a plot for an elevator pitch version of my big idea at the time. The plot was literally my plot. This confirmed my suspicions that I am a mediocre, at best, storyteller but it also made me wonder how much of what each person thinks makes them individually smart or talented is just mundane crap no better than this system can produce. Chilling to some, maybe, but confirmed my idea that what I'm doing is just for me to enjoy and that's ok.

[metanonsense]

I think it depends very much on your use case and what you expect.

My girlfriend is a judge, and sometimes we ask ChatGPT about some judicial problem for pure entertainment. To me as a law noob the answers sound absolutely convincing, but she always starts laughing and points out that the cited laws don't exist or the exemplary cases never happened.

I, on the other hand, work as a software developer and use ChatGPT as a discussion partner to get a better understanding of problem and solutions spaces. I don't expect ChatGPT to be correct but gladly take any inspiration or argument and use it to improve my own thinking process. And for this use case, I consider GPT4 absolutely invaluable. It's like a polite, knowledgeable, never busy, untirable colleague that is ready for my questions 24/7.

[spideymans]

> I, on the other hand, work as a software developer and use ChatGPT as a discussion partner to get a better understanding of problem and solutions spaces. I don't expect ChatGPT to be correct but gladly take any inspiration or argument and use it to improve my own thinking process. And for this use case, I consider GPT4 absolutely invaluable. It's like a polite, knowledgeable, never busy, untirable colleague that is ready for my questions 24/7.

I often use ChatGPT as a starting point when researching new topics (usually in the software space). In the paragraphs of lies it generates, there are usually a few keywords you can put into Google to find accurate and reliable information.

[cyanydeez]

Google is working to eradicate that.

[_nalply]

Same. More than half is bullshit but it's a form of rubber duck debugging and very inspiring to find a solution yourself.

[teknopaul]

Lies is a feature not a bug, they are open about that. You have to pick a link from search results too.

I think use is down due people going from "wow that's amazing it's even possible" to "but significantly worse than a little human effort and an Internet connected computer"

[The_Colonel]

If it was a feature, then competing LLMs could simply leave out this "feature" and take the spot of the leading AI. But the other LLMs have that problem as well.

> but significantly worse than a little human effort and an Internet connected computer

In some cases perhaps, but there are a lot of cases where asking ChatGPT and then verifying its answer is (much) faster than trying to figure out the answer on your own.

[cyanydeez]

Google used to not make up links.

[wouldbecouldbe]

I stopped using it. Then 4 came and then I used it extensively. And paid for it. Now last 2 months it's not created something useful that didn't cost me time in the end.

[reaperducer]

it quickly became apparent that ChatGPT was exceedingly willing to outright lie to you.

I've sometimes thought that these "AI" chat systems might be better if they were taught the simple human phrase "I don't know."

I think it would improve trust in these system if people knew that they had limits, and were aware of them.

It would be even better if something like the one on Bing, for example, could respond "I don't know, but here are some links to places where you might find the answer…"

It's like when it starts to become apparent that the new kid at school is a compulsive liar. Eventually people stop listening to him.

[rsynnott]

That's not really how it works, though. The line `if(dont_know) {make_shit_up()}` does not appear in the source code. It doesn't know it's lying; it doesn't know anything!

You could train one to appear terribly uncertain, but it would still 'lie'.

[AnimalMuppet]

OK, but it's probability-based. It will pick lower probability answers depending on the temperature. But that means it knows what the probabilities are.

So what it needs is a line of the form 'if(highest_probability < threshold){dont_know()}.

[cyanydeez]

That'd take a whole different model. Remember, these are billions of vectors.

[insin]

> I suppose it’s also useful for generating things like letters (things where exact truthiness doesn’t really matter). But even for that use case, I find ChatGPT creates overly verbose corporate gobbledygook whenever I ask it to generate text

This makes it perfect for creating purely time-wasting "content" - I've started sending back generated responses to people who cold-email offers to buy one of my browser extensions.

8 paragraphs (why is it always 8?) of leading waffle which is relevant to their original email. If they read more than 2 of them they've wasted more time than I spent replying.

[cyanydeez]

8 paragraphs is the context limit.