Maybe this is an unpopular take, but why are people getting so hung up on #chatGPT if it's still getting so many things wrong in an at times hard to verify way?
Maybe my queries have just been too specific but I have never actually seen the thing say something halfway useful yet. It all sounds authoritative enough, but doesn't hold up to the smallest amount of scrutiny. Am I just asking it the wrong things? Does it get better with GPT4? I just don't see it yet. So far, what I see is another Facebook that tricks people into thinking they're smart by parrotting something the internet said without fact-checking, just on demand now rather than on the schedule of others. Is it actually useful to folks yet as a way to learn new facts/information or does it just summarize and generate clickbaity blog posts so far?
@KaraLG84@zersiax I think it's a mistake to consider GPT based chat bots or whatever service as just a delivery mechanism for facts. In fact, this is the least appealing thing about it. I've been using (and comparing) various services for productivity gains. Providing exec summaries for long documents, serving as a sounding board for my writing, coming up with Excel formulas faster than my brain can manage, helping with code reviews, etc. LLMs can save hours.
@KaraLG84@zersiax I'm not suggesting that these services are the be all and end all. They have significant problems. Unless you're using them with the full understanding of their limits, it's likely that they'll fail utterly. And, if you consider that they're there to potentially make your domain-specific work more efficient, they're far more useful.
@ppatel@KaraLG84 Summarizing, I can see. Excel formulas, I can also see. A sounding board already makes me worry about any writing's end product depending on what you mean by sounding board, though. The thing just makes up too much when it doesn't have an answer ready, gives no indication that is what it's doing, and also gives no references for whatever it's giving you. Bing Chat does this a little better by at least giving you references so you can go check its work, but at that point you might as well do it all yourself :)
@zersiax@KaraLG84 As to writing, it's my policy never to use what it gives you verbatim. A GPT based system can help establish a tone or rewrite something you already have in a different way so that it doesn't sound harsh for example. I'm an excellent writer. So, it gives me efficiency. I don't want it to give me anything significant without me editing or rewriting. As to bing and search, summarization concepts apply.
@ppatel@zersiax@KaraLG84 I had both Bard and Chat GPT rewrite something I wanted to use for a video, and Bard rewrote it so badly that it looked like something entirely different. Chat GPT wasn't so bad, but both were unusable. Getting Chat GPT to write short stories about random things however, that's fine. No facts needed in that case, as it's fiction. Depending on use-case, these things can work, and really shouldn't be dismissed out-of-hand, just because at this very moment, you haven't find a direct use for it. A time may come when you do. You = general you, not you = reading this right now.
@FreakyFwoof@zersiax@KaraLG84 Oh bard is pretty terrible in general for most things. I might need to do a serious comparison of its capabilities when it comes to doing code reviews with Github's co-pilot and GPT4 now that Google has incorporated some programming capabilities. Bard can be good at some summarization tasks--if you pass on a URL for instance. I'm also reluctant to give Google more data.
@ppatel@FreakyFwoof@zersiax@KaraLG84 This might be an unpopular opinion but I wonder if people are expecting too much from something that is still so new.
I understand about incorrect answers and all that kind of stuff and they all definitely need work but the way some people go on and I am not referring to any of you here by the way, it is expected to do everything and know everything and never get things wrong.
It's machine learning.
@TheVoiceGuy@ppatel@zersiax@KaraLG84 Certainly feels like that at times. It's like expecting a five year old to know higher-level mathematics, and getting really pissed off when they don't.
@FreakyFwoof@ppatel@zersiax@KaraLG84 I know it sounds arrogant but what do people expect? These are models of pre-trained text data that are sucked in from the Internet into these big machine algorithms.
It’s going to screw up and it probably always will to some degree.
You can’t win because if it got every single thing correct, people would be freaking out and saying that it's too much that it's human in nature, it's sentient.
@TheVoiceGuy@ppatel@zersiax@KaraLG84 Someone in a youtube video put it very well, and it stuck with me. They said, and I quote: 'AI is the worse it will ever be.' Think about that for just a moment. Right now, as good as some is, and as terrible as others can be, it is the worse it will ever be. It will only get better from here. People dismissing it right now are not forward-thinking enough.
@FreakyFwoof@ppatel@zersiax@KaraLG84 Precisely!. I’m a voice actor and I have no doubt that AI is going to eat in to what is already a very hard industry but then I think well, I can clone my voice professionally through something like #elevenlabs or other similar ventures and people can pay for the use of it for commercial projects. It's already happening. I say in brace the technology. Yes there are ethical implications, of course they are.
When television came to Australia only in 1956, some people argued that it was going to destroy the moral compass of Australian youth and it should be blocked every term for God sake.
@TheVoiceGuy@FreakyFwoof@ppatel@KaraLG84 at this point I'd like to indicate that in my particular post, I was only referring to chatGPT itself. More specialized AI projects do appear quite a bit more substantial to me, elevenlabs is a good example of that
@TheVoiceGuy@zersiax@ppatel@KaraLG84 I maintain mine for exactly that reason. It's fascinating, and useful. I'm not a very good reader, not one of these super-fast braille readers, neither can I listen to speech and have it read back, but I can give Eleven Labs my voice and have it read things in a similar manner to myself. This will be very useful for my videos going forward, when I have chunks of text that need to be voiced. That's worth my $$$ personally.
@FreakyFwoof@zersiax@ppatel@KaraLG84 I have every intention of getting into my little voice over booth and recording half an hour of audio for the professional cloning for that reason.
@zersiax@FreakyFwoof@ppatel@KaraLG84 I could be wrong but definitely if you want to be a partner as a voice actor, they are still saying half an hour. I am happy to stand corrected however.
@TheVoiceGuy@zersiax@FreakyFwoof@KaraLG84 My theory is to test and embrace new tech while making all efforts to 1) try and get it accessible for us disabled people; 2) work with people to address ethical concerns; 3) see if the new tech will get me to be more effective at my work; and 4) see if I can be effective at communicating about possible efficiencies about the tools.
@TheVoiceGuy@zersiax@FreakyFwoof@KaraLG84 I agree with a lot of points in this thread so I won't repeat them. But tools are tools and these new ones are going to be everywhere. We're just at the start. It's hard to see if and when we're in the middle of a renaissance. I think we are.
@FreakyFwoof@zersiax@KaraLG84 When it comes to most of these tasks, I find Anthropic's Claude is much better than either ChatGPT or Bard. When it comes to rewriting, prompts with as many specifics you can give it, the better it is. If you find something bad, improving it with further prompting makes it better.
@devinprater@FreakyFwoof@zersiax@KaraLG84 If you're looking for another LLM based chat playground, check out the newly released chat bot available at Huggingface. The bot is based on the data compiled and labeled by volunteers with the AI Assistant project. The dataset isn't as large as the one used by OpenAI. But, it's another point of comparison.
@devinprater@FreakyFwoof@zersiax@KaraLG84 You can try signing up for Claude. But, they're not making it generally available. There is a console. But they're focused more on the API.
@zersiax I've found the same, it starts making stuff up really fast. I can see the draw, it can do things like introduce you to a tech, say CSS. And unlike docs or a video you can discuss your understanding, ask it for examples or to expand on something. But so dangerous when it starts fabricating
@joelanman right but even with that its knowledge ends in 2021, so ask it about React and you'll just get out of date archaisms now :) I haven't really tried making it generate code, maybe it's a star at that but on the whole ...yeah no, not enthused :)
Add comment