|
Boris Galerkin posted:If ChatGPT generates text randomly then what use is it to me, the person doing a peer review of the paper that's citing information out of ChatGPT? Or me, the person reading this article (that cites information out of ChatGPT) trying to get information to use in my own research? If you had direct access to the model you could do this! You would need all of the backend info OpenAI baked into ChatGPT, but with local LLMs, you can precisely replicate responses based on specific input and parameters.
|
#
?
Jun 11, 2023 17:42
|
|
|
|
| # ? Jun 8, 2024 07:08 |
|
|
Main Paineframe posted:The example is written the way it is as a response to the limitations and issues in citing ChatGPT Yes that's what I'm saying. The article describes the limited context in which citing ChatGPT would make sense because of its many limitations, errors, and general weirdness, and it prescribes how you should cite it (which is the same format whether your use of it makes sense or not). That's why the "lmao the APA style idiots should've just said never to cite ChatGPT" reaction is stupid.
|
|
#
?
Jun 11, 2023 17:46
|
|
|
This is all of the input data supplied to an LLM when running inference to generate results from a trained model. These things start feeling a lot less magical when you can prod at them to see how they work. If nothing changes between hitting generate, you'll get the same response 100% of the time, as would anyone else running with these same inputs, including seed, on the same model.  
|
|
#
?
Jun 11, 2023 17:58
|
|
|
LASER BEAM DREAM posted:This is all of the input data supplied to an LLM when running inference to generate results from a trained model. These things start feeling a lot less magical when you can prod at them to see how they work. If nothing changes between hitting generate, you'll get the same response 100% of the time, as would anyone else running with these same inputs, including seed, on the same model. Well, no, because this set of inputs includes a randomized seed. Someone entering the exact values listed here would get a different seed from what you did, and so potentially a different response. In order to have it consistently reproducible, you would also have to provide the exact seed you ended up with.
|
|
#
?
Jun 11, 2023 18:00
|
|
|
Thinking about it more, and with seeing those screenshots above, and taking into consideration of a paper researching LLMs might want to cite LLM output, I think the problem for me is the word “reference.” Like, say I write a journal article about the results of my numerical model, citing the results as a reference is not a thing that people do. Whether I use commercial software like ANSYS or custom in-house code, nobody cites the model as a reference and doing so makes no sense at all. What you might do is cite other sources as reference to the model you’re running, and you may cite other sources to reference the model parameters you chose to use, but citing the software that spits out the numbers? That’s just not a thing. Or put it this way, if I were to write a paper on LLM I could see it making perfect sense to include a table showing all those parameters displayed in the pictures above. I could see it making perfect sense to cite OpenAI’s paper(s) on their LLM as reference explaining the models. It would make sense to include input and output given to ChatGPT as well. But it makes NO SENSE to cite ChatGPT as a reference here. Or even simpler: if you do a math problem on a calculator it makes no sense to cite Casio or Texas Instruments as a reference on how you got the output “69”. Boris Galerkin fucked around with this message at 18:33 on Jun 11, 2023 |
|
#
?
Jun 11, 2023 18:31
|
|
|
Main Paineframe posted:Well, no, because this set of inputs includes a randomized seed. Someone entering the exact values listed here would get a different seed from what you did, and so potentially a different response. In order to have it consistently reproducible, you would also have to provide the exact seed you ended up with. I specified seed, but you're correct. You would want to include this as well. 
|
|
#
?
Jun 11, 2023 18:40
|
|
|
Is it really that much sillier and non-reproducible than "Goon McGoonerson, Personal Correspondence"? It's not like people expect to be able to demand the personal correspondence to corroborate, right?
|
|
#
?
Jun 11, 2023 20:53
|
|
|
Absurd Alhazred posted:Is it really that much sillier and non-reproducible than "Goon McGoonerson, Personal Correspondence"? It's not like people expect to be able to demand the personal correspondence to corroborate, right? Yes, there is a difference.
|
|
#
?
Jun 11, 2023 21:48
|
|
|
Absurd Alhazred posted:Is it really that much sillier and non-reproducible than "Goon McGoonerson, Personal Correspondence"? It's not like people expect to be able to demand the personal correspondence to corroborate, right? No, although writers are highly encouraged to include the part of the correspondence they're relying on as a quote. However, with a well-cited personal correspondence, it's entirely possible that someone else can write to the person and ask "hey, X said in their paper that you told them this, are they full of poo poo?". That said, it's not like "personal correspondence" citations are highly encouraged either. We can look at the APA rules for personal correspondence citations and see exactly how highly they think of that: quote:Works that cannot be recovered by readers are cited in the text as personal communications. Personal communications include emails, text messages, online chats or direct messages, personal interviews, telephone conversations, live speeches, nonarchived social media livestreams (e.g., Instagram Live, Twitter Spaces), unrecorded webinars, unrecorded classroom lectures, memos, letters, messages from nonarchived discussion groups or online bulletin boards, and so on. The clear implication is that "personal correspondence" is a citation of last resort, to be used when the information has essentially never been written down or recorded anywhere outside of your personal email inbox.
|
|
#
?
Jun 11, 2023 22:57
|
|
|
Reproducible psudorandom beats this poo poo out of true random for just about anything except deployed cryptography.
|
|
#
?
Jun 11, 2023 23:12
|
|
|
Remulak posted:Reproducible psudorandom beats this poo poo out of true random for just about anything except deployed cryptography. i keep a dice box on my desk and roll a die when i need something. it's cheap and simple and quick.
|
|
#
?
Jun 11, 2023 23:19
|
|
|
Arivia posted:i keep a dice box on my desk and roll a die when i need something. it's cheap and simple and quick. Citation required
|
|
#
?
Jun 11, 2023 23:19
|
|
|
Boris Galerkin posted:Citation required Arivia [Arivia]. (2023, June 11). Tech Nightmares 6: entries by user "Nword Balls" that show a pretty convincing goatse [Online forum post]. Something Awful. https://forums.somethingawful.com/showthread.php?noseen=0&threadid=3763277&perpage=40&pagenumber=1362#post532447332 Arivia fucked around with this message at 03:29 on Jun 12, 2023 |
|
#
?
Jun 11, 2023 23:23
|
|
|
I don’t really see the problem with having a citation style for ChatGPT. It’s obviously not an endorsement for the technology. And the irreproducibility is something to be addresssed in the text, just like any other methodological issues. Don’t attack a style guide just because you don’t like the source.
|
|
#
?
Jun 11, 2023 23:25
|
|
|
Academics are mostly going to cite ChatGPT either as an example of an LLM or as a comparison model in an experiment. BERT and GPT-3 papers have tens of thousands of citations so there's nothing strange at all about citing entire models, but the difference with ChatGPT is there's no similar foundation paper.
|
|
#
?
Jun 12, 2023 00:47
|
|
|
https://twitter.com/m_d_c_t/status/1668070983498977280 Ahhhh hell, another data point in favor of 'AI nattering is a red flag for management eptitude'.
|
|
#
?
Jun 12, 2023 18:58
|
|
|
Discord is on fire?
|
|
#
?
Jun 13, 2023 07:24
|
|
|
Good?
|
|
#
?
Jun 13, 2023 07:30
|
|
|
So Discord is on fire, Reddit is going through open revolt, twitters dying, facebook's almost dead to the under 40's. Is there any social media that isn't in a decline at the moment? I assume tik toks doing just fine?
|
|
#
?
Jun 13, 2023 07:37
|
|
|
Besidies some issues with Government intervention, tik tok is making money hand over first.
|
|
#
?
Jun 13, 2023 07:42
|
|
|
dr_rat posted:Is there any social media that isn't in a decline at the moment? Buddy you're on one.
|
|
#
?
Jun 13, 2023 08:00
|
|
|
Tayter Swift posted:Buddy you're on one. Hoorah, twenty year old decisions well made!
|
|
#
?
Jun 13, 2023 08:03
|
|
|
One weird social media trick they don’t want you to know: ruthlessly banning assholes!
|
|
#
?
Jun 13, 2023 08:19
|
|
|
dr_rat posted:So Discord is on fire, Reddit is going through open revolt, twitters dying, facebook's almost dead to the under 40's. Is there any social media that isn't in a decline at the moment? Mega Comrade posted:Besidies some issues with Government intervention, tik tok is making money hand over first. Too bad
|
|
#
?
Jun 13, 2023 08:20
|
|
|
YouTube is still great. Educational YouTube is legit. I spent a recent flight watching downloaded videos about religious history. Then at home I watch everything from Succession scenes to cooking videos to gadget reviews to hotel room reviews. No bullshit. Premium is a clean no-frills product with a clear value proposition too. That said I’m really bummed about the recent reddit shutdowns because it literally makes it like 200% harder to find honest, good reviews of things. YouTube only kinda covers that gap. I wish Reddit would sack up and do the right thing.
|
|
#
?
Jun 13, 2023 08:57
|
|
|
What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say.
|
|
#
?
Jun 13, 2023 10:40
|
|
|
They keep trying to make it be something other than voicechat / IRC with media hosting
|
|
#
?
Jun 13, 2023 10:55
|
|
|
Boris Galerkin posted:What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say. I hate all technology, even the ones keeping me alive. Especially those.
|
|
#
?
Jun 13, 2023 11:16
|
|
|
Boris Galerkin posted:What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say. The quality of their app deteriorates quickly, too. I had to uninstall it from my phone because opening the message from notifications made all the channels and contacts display only the said message, requiring to force close the app and reopen
|
|
#
?
Jun 13, 2023 11:46
|
|
|
Boris Galerkin posted:What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say. As someone who uses discord and has done for years, it is slack for video games. I just wonder when the VC money is gonna dry up cos no way they are covering costs with nitro.
|
|
#
?
Jun 13, 2023 12:02
|
|
|
Boris Galerkin posted:What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say. StratGoatCom posted:https://twitter.com/m_d_c_t/status/1668070983498977280
|
|
#
?
Jun 13, 2023 14:01
|
|
|
Don't worry, I just took their user feedback survey and was very frank about how they're about to gently caress things up. I'm sure it'll get them turned around in a jiffy
|
|
#
?
Jun 13, 2023 14:36
|
|
|
Mega Comrade posted:As someone who uses discord and has done for years, it is slack for video games. Didn't Sony acquire Discord?
|
|
#
?
Jun 13, 2023 14:48
|
|
|
Mister Facetious posted:Didn't Sony acquire Discord? No they have a minority stake in the company
|
|
#
?
Jun 13, 2023 14:52
|
|
|
Boris Galerkin posted:What’s wrong with Discord? Are we hating on it now because “it’s a slack but for teenagers who play video games”? And if that’s what you think then “ok boomer” is all I’ve got to say. Remember when Discord tried to launch a video game storefront?
|
|
#
?
Jun 13, 2023 15:29
|
|
|
It's trying to turn into regular slack
|
|
|
#
?
Jun 13, 2023 15:48
|
|
|
Watermelon Daiquiri posted:It's trying to turn into regular slack They should try turning it into Tic tok. That's still popular just do that! Do the kid's these days still uses those tamagotchi things? Add something like that in as well!
|
|
#
?
Jun 13, 2023 15:55
|
|
|
Watermelon Daiquiri posted:It's trying to turn into regular slack It's trying to turn into loving facebook/twitter. The whole stupid movement towards removing the old decimator system for usernames was a sign of wanting their users to have true unique identifiers to allow for it to be an open platform where they can sell special usernames to companies. Can't have Nabisco competing with nabisco#0001-9999.
|
|
#
?
Jun 13, 2023 15:58
|
|
|
Crain posted:It's trying to turn into loving facebook/twitter. Well they hosed that up already: 
|
|
#
?
Jun 13, 2023 16:00
|
|
|
|
| # ? Jun 8, 2024 07:08 |
|
|
That's weird because my Discord is just a bunch of chat rooms. There's no Facebook/Twitter like functionality I see anywhere. I guess they are A/B testing and I'm not in the cohort getting these weird features.
|
|
#
?
Jun 13, 2023 16:10
|
|