|
Really interesting video I came across. "The AI revolution is rotten to the Core" https://www.youtube.com/watch?v=-MUEXGaxFDA At the end he talks about something I'm more concerned about, that is people befriending the chatbot.
|
#
?
Oct 6, 2023 15:19
|
|
|
|
| # ? May 5, 2024 18:17 |
|
|
Tarkus posted:At the end he talks about something I'm more concerned about, that is people befriending the chatbot. At least at the moment it would be like befriending someone with moderate dementia. The context awareness of GPT-4 is impressive but the responses start falling apart if you keep them going for too long. This forces you to start a new chat, erasing all of its memories. Once they manage to give the chat-bots access to some kind of permanent editable database this could make things even crazier.
|
|
#
?
Oct 6, 2023 19:45
|
|
|
hbag posted:i have an artificial intelligence class this semester and the lecturer is encouraging us to ask chatgpt poo poo and im not sure how to break it to him that it just makes poo poo up and is not at all a reliable source I'm guessing the purpose isn't that he thinks it's reliable but to see how the manner in which you ask things affects the response, how you can game it to get the output you want, how it could provide misleading information, etc. I would hope so at least.
|
|
#
?
Oct 6, 2023 20:12
|
|
|
Mescal posted:What benefit is there to asserting current weakness in the machine? Of course it's a weak thinker! We've given it a highly impoverished existence, yet it learns and creates. It can't experience time and it has no body, yet it thinks. Assuming you aren't being sarcastic, so people don't go and rely on it for accuracy in anything where it could potentially cause harm. c.f. every single company that has replaced human beings with ChatGPT engines for anything. Too soon to drink the Kool-aid there, people. If you are being sarcastic, then it is to assert our rightful human dominance in the world.
|
|
#
?
Oct 6, 2023 20:48
|
|
|
Mozi posted:I'm guessing the purpose isn't that he thinks it's reliable but to see how the manner in which you ask things affects the response, how you can game it to get the output you want, how it could provide misleading information, etc. the guy leading the computer science course also said we can use chatgpt for research and cite it. although if we just copy paste poo poo from it thats contract cheating which leads to immediate expulsion
|
|
#
?
Oct 6, 2023 20:51
|
|
|
Citing ChatGPT seems... very suspect to me, if the assumption is that it's giving you a factual answer. Even if it is correct I still feel like an actual source would be good.
|
|
#
?
Oct 6, 2023 20:57
|
|
|
How on earth would you cite ChatGPT? It's output, even when correct, changes every time you use it. It's not a document nor is it published in any way.
|
|
#
?
Oct 6, 2023 21:03
|
|
|
I can't imagine it being a different from citing people. You wouldn't cite an email, even if it came from an expert. You cite publications and official documents etc
|
|
#
?
Oct 6, 2023 21:09
|
|
|
hbag posted:the guy leading the computer science course also said we can use chatgpt for research and cite it. although if we just copy paste poo poo from it thats contract cheating which leads to immediate expulsion This is dogshit pedagogy.
|
|
#
?
Oct 6, 2023 21:24
|
|
|
This popped up the other day. People have been speculating about using AI to confuse the public in relation to politicians https://www.bbc.com/news/world-africa-66987869.amp quote:AI: Voice cloning tech emerges in Sudan civil war quote heavily edited to shrink down the article, much longer in link
|
|
#
?
Oct 6, 2023 21:46
|
|
|
In that vein, does anyone remember that time back after 9/11 when there was a video of a supposed Bin Laden admitting to doing everything but it was shot only from the back, he was much fatter and didn't sound like Bin Laden? Am I imagining this? That the US government released such a fake video?
|
|
#
?
Oct 6, 2023 21:49
|
|
|
Tarkus posted:How on earth would you cite ChatGPT? It's output, even when correct, changes every time you use it. It's not a document nor is it published in any way. i think they said something about putting its output in an appendix with your references. idk. either way im not using it at all lmao
|
|
#
?
Oct 6, 2023 21:54
|
|
|
Doctor Zero posted:Assuming you aren't being sarcastic, so people don't go and rely on it for accuracy in anything where it could potentially cause harm. c.f. every single company that has replaced human beings with ChatGPT engines for anything. Too soon to drink the Kool-aid there, people. Surely you don't take what I was saying as an endorsement of an AI's general factuality? I agree it gives inaccurate results under countless circumstances. And NOBODY'S saying to start citing GPT, that's fantasy y'all came up with.
|
|
#
?
Oct 6, 2023 22:47
|
|
|
I think it's fine to let students cite ChatGPT, but it should be made crystal clear to them that factchecking the AI is their responsibility. Any false statements made by the AI will be interpreted as false statements made by the student and will therefore affect their grade.
|
|
#
?
Oct 6, 2023 23:03
|
|
|
at that point just cite what you used to fact check it as your reference
|
|
#
?
Oct 6, 2023 23:07
|
|
|
Let me repeat: Nobody said that you should cite GPT as a source of information! But its use attendant to the production of knowledge is salient. In matters of public record, that could be thorny. Citing the way you USED an AI (like, linking the chat) in the production of an academic or other work could be a thing that'd add great transparency to the production of knowledge. Like, imagine if Beethoven's correspondence or the Buckminster Fuller library contained their GPT logs. It'd be more fun to read than a diary. The fact that AI is now bound to the production of knowledge does raise questions of transparency, even if the "citing" word doesn't fit.
|
|
#
?
Oct 6, 2023 23:23
|
|
|
I'm not sure but I don't think anyone is saying that you're all an board for citing chatGPT. But yes, another fear or concern of mine is that a ton of our information will be filtered through the lens of these new LLM's. It's going to be the source of a great cultural shift if it finds greater widespread use.
|
|
#
?
Oct 6, 2023 23:38
|
|
|
Tarkus posted:ChatGPT is not merely a search engine. In fact, the model largely does not contain any of the original data that it was trained on (Aside from the basic tokens), only the weighted relationships that made up that data. The original text, particularly if it's common can be regurgitated because that is the highest weighted relationship but it does not contain it as a string or stored file. That is why GPT has trouble with certain things like less common facts, reverse relationships and certain elements of logic. Can it be sussed out from human text? sure, it's a large language model not a person. Nobody is saying it is a person. That said, people who state that it outputs nonsense or gibberish kind of bug me because, frankly, I feel my output is less coherent that what GPT4 puts out. I think those that claim those things are misinformed or reacting out of fear. It's like if I trained a cat to speak and people would would come by and tut-tut about the cat not using full sentences and claiming it's not talking because it was only trained to talk. It's still impressive. Wwith enough playing around with weird prompts you can get GPT to output conceptual relationships with clearly correct associations but reverse, which I find is a really interesting peek at the weirdness of how it thinks. For example, get GPT 3.5 to talk about the heights of various people and it will inevitably get 'shorter/taller' in reverse at some point.
|
|
#
?
Oct 7, 2023 01:20
|
|
|
Is there an MLA standard for citing ChatGPT yet
|
|
#
?
Oct 7, 2023 02:15
|
|
|
Hadlock posted:Is there an MLA standard for citing ChatGPT yet I don't know about MLA but there are guidelines from the APA, I'm sure they'll change though once it's put in the actual style guide. https://apastyle.apa.org/blog/how-to-cite-chatgpt
|
|
#
?
Oct 7, 2023 02:19
|
|
|
His Divine Shadow posted:Just used AI for something work related for the first time ever. Photoshop has some new AI functions and the generative fill is real impressive, got some nice photo but wrong aspect ratio, used PS to expand the canvas and fill in the blanks and the results are actually usable I'd say. Does the latest version still have the 1024 x 1024 limit? Previously anything beyond that was simply 1024x1024 stretched out.
|
|
#
?
Oct 9, 2023 14:40
|
|
|
Tears In A Vial posted:Does the latest version still have the 1024 x 1024 limit? Previously anything beyond that was simply 1024x1024 stretched out. wouldve thought itd pipe into an ai image upscaler
|
|
#
?
Oct 9, 2023 15:15
|
|
|
Mescal posted:Surely you don't take what I was saying as an endorsement of an AI's general factuality? I agree it gives inaccurate results under countless circumstances. And NOBODY'S saying to start citing GPT, that's fantasy y'all came up with. I didn’t, no
|
|
#
?
Oct 9, 2023 15:19
|
|
|
Here's what happened when I asked chatgpt for jokes about Dwarves.    
|
|
#
?
Oct 9, 2023 16:36
|
|
|
Next time, post that before the writers’ strikes, save everyone a bunch of trouble. Their jobs are clearly safe.
|
|
#
?
Oct 9, 2023 16:44
|
|
|
Has anyone tried the premium AI Dungeon recently? Last time I used the free model it was poo poo. I've been using Claude.ai to make my own little custom RPGs since it has the largest context window around. It's pretty drat good at doing text adventures, good enough to pretty much sell the illusion, but I want to have a little separate inventory window, character stats, have it automatically summarize events and goals, have standardized turns, combat that has at least some vague rules associated with it, etc. If AI Dungeon has improved significantly in the past year then I might give that a shot.
|
|
#
?
Oct 9, 2023 17:59
|
|
|
I'm in a degree program for data science which requires a lot of coding. If I use ChatGPT to help generate code, I'm required to cite the prompt and output.
|
|
#
?
Oct 10, 2023 06:55
|
|
|
credburn posted:I'm in a degree program for data science which requires a lot of coding. If I use ChatGPT to help generate code, I'm required to cite the prompt and output. This actually makes a lot of sense as it helps see into your thought process, and honestly you normally are going to iterate a few times "no this doesn't work" how much of the it do you solve yourself? "I think you mean if { not if (" And let's be honest there's a good chance something actually usable will exist at some point in your career like this that is actually usable.
|
|
#
?
Oct 10, 2023 15:31
|
|
|
pixaal posted:This actually makes a lot of sense as it helps see into your thought process, and honestly you normally are going to iterate a few times "no this doesn't work" how much of the it do you solve yourself? "I think you mean if { not if (" And let's be honest there's a good chance something actually usable will exist at some point in your career like this that is actually usable. I agree, this makes sense. I like that. One thing I've noticed when coding with GPT is that the context window is your friend. If you ask GPT to generate code and you don't fully know the end point and don't explain yourself, it'll generate a fair number of mistakes sometimes. I suspect that is because it has no real context of anything it's coding. However, if you've done some work already and you put some of the functions or even a rough meta your generated code is to interact with into the chat, the code becomes much, much better. People often forget that GPT is also great at diagnosing code too, particularly if you include an error or unwanted behavior and give the suspected code along with it. It has spotted things, even in embedded applications, that would've gone right over my head and would've had me ripping my hair out. The one big thing holding it back from being super-useful is its small context window. I think standard GPT4 is 8k tokens, GPT 3.5 has a 16k version available through the API. There is a model for GPT4 at 32k tokens but I can't use it, probably only for corporate customers.
|
|
#
?
Oct 10, 2023 16:20
|
|
|
GPT4 + Dalle3 is at least fun for "this discord poster has this avatar" and just describe the avatars then throw some quotes from discord and watch the magic happen. Plenty of tokens for that.
|
|
#
?
Oct 10, 2023 17:44
|
|
|
Speaking of which, how do we know if we have access to DallE-3 in GPT plus? I have GPT plus and don't see anything and I'm all out of bing boosts. I'd pay money to be able to generate some more images.
|
|
#
?
Oct 10, 2023 17:51
|
|
|
Tarkus posted:Speaking of which, how do we know if we have access to DallE-3 in GPT plus? I have GPT plus and don't see anything and I'm all out of bing boosts. I'd pay money to be able to generate some more images. Open a new window, if you have it it is a dropdown under the model select
|
|
#
?
Oct 10, 2023 17:58
|
|
|
Tunicate posted:Open a new window, if you have it it is a dropdown under the model select Ah, yeah. Not yet for me. 
|
|
#
?
Oct 10, 2023 18:00
|
|
|
Hover over this lil guy efb
|
|
#
?
Oct 10, 2023 18:01
|
|
|
Tarkus posted:Ah, yeah. Not yet for me. It only shows on the website for me, but on mobile if I go into history I can still generate new images it just acts like GPT is talking instead of showing image generating spinning and the prompts. There's no way to browse them as a set you have to click each image on mobile and no way to see the prompt without asking for it. (you can of course hop on web and check in that same chat session)
|
|
#
?
Oct 10, 2023 18:13
|
|
|
Yeah, it shows up on nothing. Could be because I'm from Canada. I know google isn't allowing bard in Canada right now because of a tiff they're having.
|
|
#
?
Oct 10, 2023 19:09
|
|
|
Tarkus posted:Yeah, it shows up on nothing. Could be because I'm from Canada. I know google isn't allowing bard in Canada right now because of a tiff they're having. I am in the USA and also dont have DALL E in my plugins, despite having beat features enabled.
|
|
#
?
Oct 10, 2023 19:10
|
|
|
So I now have the GPT-V, The Dall-E3 and now the speech/chat on mobile. Has anyone tried it (the speech) yet? It's.... interesting.
|
|
#
?
Oct 19, 2023 04:41
|
|
|
Tarkus posted:So I now have the GPT-V, The Dall-E3 and now the speech/chat on mobile. Has anyone tried it (the speech) yet? It's.... interesting. I didn’t turn it on as I’m not sure what’s is use it for. What do you use it for that you couldn’t just use voice to text for?
|
|
#
?
Oct 19, 2023 12:08
|
|
|
|
| # ? May 5, 2024 18:17 |
|
|
CarForumPoster posted:I didn’t turn it on as I’m not sure what’s is use it for. What do you use it for that you couldn’t just use voice to text for? It will actually carry on a dialog with you. It's not great... yet. But that kind of stuff concerns me.
|
|
#
?
Oct 19, 2023 14:20
|
|