This wasn't an intentional feature; they're actually trying to train it with fine-tuning to add this as an ability. It's one area that highlights the difference between it imitating the text it's been seeing, instead of actually understanding what it's saying -- since most of its training data is of the form "(ask a question) (response to question)" overwhelmingly more often than "(ask a question) (say you don't know, the end)", it is trying to be a good imitator and do the same, and come up with some plausible nonsense even if it doesn't know the answer.
And sometimes that's exactly what I want, too. I use LLMs like ChatGPT when brainstorming and fleshing out fictional scenarios for tabletop roleplaying games, for example, and in those situations coming up with plausible nonsense is specifically the job at hand. I wouldn't want to go "ChatGPT, I need a description of the interior of a wizard's tower is like" and get the response "I don't know what the interior of a wizard's tower is like."
At one point I messed around with a lore generator that would chop up sections of "The Dungeon Alphabet" and "Fire on the Velvet Horizon" along with some other stuff, and feed random sections of them into the LLM for inspiration and then ask it to lay out a little map, and it pretty reliably came up with all kind of badass stuff.
Part of the problem is fine tuning is very shallow, and that a contributing issue for claiming to be right when it isn't is the pretraining on a bunch of training data of people online claiming to be right when they aren't.
Yeah. It is fairly weird to me that it’s such a common thing to do to take the raw output of the LLM and send that to the user, and to try use fine-tuning to get that raw output to look some way that you want.
To me it is obvious that something like having the LLM emit a little JSON block which includes some field which covers “how sure are you that this is actually true” or something, is more flexible and simpler and cheaper and works better.
I haven’t really had a problem with it… maybe like 5% of the time it will want to do something a little bit weird like wrapping it in ``` but in general it seems like it works well enough to be able to parse with a program and just retry if it does something weird.
You do have to set it up a little carefully, I guess - like usually I’ll give it an example of what I want it to emit, and that’ll be good enough that that’s the form it will follow when it’s emitting stuff back to me. But yeah if you give it prompting and a specific machine readable thing to give back that seems like it usually works better than sticking with English and hoping it goes “yes” or “no” or etc like that.
But seriously, you can try it too. Download the LLaMA 3 model using the Ollama program. I am using Open WebUI as a frontend. I got this answer most of the times I asked.
edit: Wow, you are right, there's a spelling error in that answer. Rare event, I don't see this every day.
But what is missing is a more generalized infrastructure for detecting content ownership and providing compensation in a general purpose way. This is one of the great business opportunities of the next few years, awaiting the kind of breakthrough that pay-per-click search advertising brought to the World Wide Web.
There are people out there who looked at the World Wide Web and immediately thought: "how can I use this for advertising?" From that type of perspective, this is a well-thought-out summary of AI-related copyright issues.
Sounds like a shitty implementation, it adds an additional prompt everytime you ask something that you have to confirm if you also want it to lookup with ChatGPT and then it just looks it up via text? Kind of wish it was just tied to an extra command, like “Hey Siri, check with ChatGPT…” and then it spoke the thing out. Otherwise, if I just wanted to read and write prompts, I’m just going to type stuff out directly with ChatGPT, without talking to Siri in the first place. The only benefits Siri has is voice communication and telling my kids stupid knock-knock jokes on command, they’re not really adding anything to it.
I actually decided to cancel my ChatGPT subscription since it started to be so useless, for code generation, and writing help.
I'm so far pretty happy with Claude, but I've only used it Friday.
Like one of the things that it would do is give me wrong code, I'd fix it, give it back the corrected code to add something else, and it would remove the corrections and other things it added earlier!
Yeah. Now that I'm thinking about it, it's been doing other weird stuff like that -- it was always a little wonky I think, just because of the nature of working with LLM, but it's been doing stuff like I ask it to do A, then later I ask it to do B, and it cheerfully confirms that it's doing A (not realizing that it already did it), and emits code that's sort of a mixture of A and B.
IDK. I've also heard good things about Mistral. I just tried to create a Claude account but the phone verification isn't working and I have no idea why. I may check it out though; if this is accurate then it's pretty fuckin fancy and the Haiku model is significantly cheaper and smarter even than the 3.5 API which has a notable lack of cleverness sometimes.
ChatGPT has been doing this thing where I’ll ask it to do A, B, C in sequential, iterative prompts, but when it does C, it removes the lines it added for B. Then when you tell it that it removed B and needs to add it back in, it undoes C while saying it’s doing A, B, C. So frustrating.
ChatGPT
Newest