What do people actually use ChatGPT for? OpenAI provides some numbers. (arstechnica.com)
from sabreW4K3@lazysoci.al to technology@beehaw.org on 16 Sep 07:08
https://lazysoci.al/post/34099277

#technology

threaded - newest

catharso@discuss.tchncs.de on 16 Sep 07:25 collapse

<img alt="" src="https://discuss.tchncs.de/pictrs/image/c2323821-0e8c-4bb9-898b-0c7d285a0e34.png">

spechter@feddit.org on 16 Sep 07:53 collapse

So almost 50% use it as a more energy intensive search engine, huh?

Perspectivist@feddit.uk on 16 Sep 09:22 next collapse

You can ask it much more complex questions than you can google and you can ask follow-up questions too.

jherazob@beehaw.org on 16 Sep 09:28 collapse

You just have zero guarantee that the answers will be real

Perspectivist@feddit.uk on 16 Sep 09:43 next collapse

No, but you can see if the answer makes sense and then fact check it using Google if you need to. Which still doesn’t give you 100% gurantee either.

<img alt="" src="https://feddit.uk/pictrs/image/320be1cb-5a3f-494c-a511-dc4153226fb4.webp">

Chozo@fedia.io on 16 Sep 09:48 next collapse

For what it's worth, ChatGPT has gotten better at citing its sources, so it's easier to fact-check it.

rozodru@piefed.social on 16 Sep 11:42 collapse

it's true that it has gotten better with sources. However remember the context of the conversation? much worse. But I can see the direction OpenAI is trying to take it. short one off responses/solutions with little followup.

It is better than Claude though. Claude will just make stuff up or say EVERYTHING is a "known issue" when it isn't.

stray@pawb.social on 16 Sep 10:15 collapse

That’s also true of traditional searches because the resulting webpages can just be whatever bullshit someone wrote. It will only be true that they said it. You still have to use your brain to assess the trustworthiness of the info.

megopie@beehaw.org on 16 Sep 11:55 collapse

They get things wrong at a far higher rate than most of the websites that tend to end up at the top of a web result, and they get things wrong in weird ways that won’t stand out to users in the same way a shitty website will. These probabilistic text generators are much better at seeming like they have the correct answer than actually providing it.

Hirom@beehaw.org on 16 Sep 09:26 next collapse

Yep, using ChatGPT is a way to increase one’s environmental footprint.

And the energy cost doesn’t appear to be fully passed to users yet, as OpenAI isn’t profitable yet. There are even free LLM services. So users don’t have an insentive to prefer less polluting alternatives, such as classic search engines.

megopie@beehaw.org on 16 Sep 11:17 collapse

It’s crazy how much money they are losing, and that’s with most of their compute being provided by Microsoft at cost, if not for free in exchange for the use of their models in Microsoft products.

Both they and Anthropic talk about their business as if they’re a software as a service company, but most SAS doesn’t get more expensive to run the more users there are, not to mention their conversion rate of free users to payed users is abysmal. Like, it’s an unsalvageable train wreck of a business model, I don’t see ether surviving more than a year unless they radically change their business models.

DarkAri@lemmy.blahaj.zone on 16 Sep 21:03 collapse

Google really sucks now for some reason. It used to be much better.