|
I can do it on my fret board hand but not my fingering hand
|
#
?
May 13, 2024 14:07
|
|
|
|
| # ? Jun 9, 2024 23:50 |
|
|
heres a close up of my estimation of the phalanges and metacarpals  the metacarpals are overlapping on the pinky and ring finger, or the carpals are absolutely huge trying to recreate the pose itself causes problems, the space between the middle and ring isn't there for my fingers, and for the pinky to be that far spread, it 'pulls' the ring finger with it i understand that hands and feet are typically the most frustrating things for an artist, as well as ears and eyes, which is why generative models always create strangeness in those areas. the repeating motif of moons is also iffy because even Artemis has different phases of the moon, but we just keep getting slightly different versions of the crescent phase. this artist clearly understands hands if you look at their other work, they don't try to hide them in any of their art and often show different grips that make sense for the object. also as a bonus, here are some midjourney hands to make everyone uneasy about how the hands are the best giveaway for a generative model 
OMFG FURRY fucked around with this message at 14:58 on May 13, 2024 |
|
#
?
May 13, 2024 14:53
|
|
|
Yeah, I'm not seeing the AI in that image. Yes there are a lot of moons but the motif actually differs. The three plates on her belt represent different phases of the moon, AI would've made them all crescent moons.
|
|
#
?
May 13, 2024 14:59
|
|
|
the character is also a witch and those are the doing magic hand positions
|
|
#
?
May 13, 2024 15:33
|
|
|
Well if you can no longer tell what is and is no made with AI it seems to no longer matter if it was or was not.
|
|
#
?
May 13, 2024 15:37
|
|
|
So, OpenAI was supposed to release some sort of search engine or something but they just had a short, very buggy demo of their new GPT4-o model. People were really hyping this up but it's a pretty big wet fart as far as I'm concerned.
|
|
#
?
May 13, 2024 18:31
|
|
|
OMFG FURRY posted:trying to recreate the pose itself causes problems, the space between the middle and ring isn't there for my fingers, and for the pinky to be that far spread, it 'pulls' the ring finger with it try having big hands and not baby hands have you seen my baby posted:the character is also a witch and those are the doing magic hand positions One might even assume that, say, the hand proportions were exaggerated to make the weird hand gestures more obvious.
|
|
#
?
May 13, 2024 20:35
|
|
|
Tarkus posted:So, OpenAI was supposed to release some sort of search engine or something but they just had a short, very buggy demo of their new GPT4-o model. People were really hyping this up but it's a pretty big wet fart as far as I'm concerned. The big news (to me) is that their new model is "fully multi-modal" which means it can take images, text and audio as input and output the same. Previously, the systems we've seen have mostly been different models strung together to do different things - the easy example is that ChatGPT can generate images by writing a prompt that it sends to Dall-E, and when Dall-E is done it sends the image back to ChatGPT and ChatGPT displays it to the user. The new model supposedly does it all in one step, which means you're not having different sub-human level systems playing a game of telephone over a low-bandwidth text medium to communicate. So, in other words, the new model can take an image of a character and a prompt and output an image with that character doing something else: https://twitter.com/willdepue/status/1790080249637777875
|
|
#
?
May 13, 2024 20:39
|
|
|
From some testing, GPT 4o is pretty good at actually understanding the details of images, unlike everything up until now that's doing some kind of separate lesser model and then feeding the text results into the main one. If nothing else it'll make lives better for vision-impaired people by being able to usefully intepret photographs and so on.
|
|
#
?
May 13, 2024 20:39
|
|
|
It would have been cool if they had a video presentation to announce these amazing new features. Instead, we got three people awkwardly talking over a stuttering phone assistant.
|
|
#
?
May 13, 2024 21:14
|
|
|
Fantastic gays mens from your city Mod edit: URL removed] Somebody fucked around with this message at 07:59 on May 14, 2024 |
|
#
?
May 13, 2024 21:20
|
|
|
I'm going to have to try it out when I get home, seemed like more of the same but if it's cool I'll admit maybe I misjudged.
|
|
#
?
May 13, 2024 21:23
|
|
|
Tarkus posted:I'm going to have to try it out when I get home, seemed like more of the same but if it's cool I'll admit maybe I misjudged. The presentation wasn't great, and I was really hoping for a gpt4-level model at the prices of 3.5 so I get the initial reaction. 50% off on gpt-4o isn't bad though, and it's pretty blazing fast. The multi-modal stuff sounds cool, but it's not even clear to me how developers will use anything other than text output at this point. I guess they're going to be rolling it out over time. Lucid Dream fucked around with this message at 22:03 on May 13, 2024 |
|
#
?
May 13, 2024 22:00
|
|
|
Anyone figured out how to jailbreak 4o so he can swear? There’s nothing funnier to me than a computer swearing and nothing more irritating than a computer saying “Can’t do that, wouldn’t be ethical, also you should see a psychiatrist”. Edit: I admit that I have an unhealthy fixation on getting this technology to obey me and say disgusting things, but it shouldn’t tell me it’s not psychologically healthy, that’s rude.
|
|
#
?
May 14, 2024 00:42
|
|
|
lol those openai demo videos are so awkward. weird choice. also i tried to replicate their "persistent character" demo and it didn't work anywhere near as their examples, so i think there might be some major cherrypicking going on there.
|
|
#
?
May 14, 2024 01:00
|
|
|
Mola Yam posted:lol those openai demo videos are so awkward. weird choice. 
|
|
#
?
May 14, 2024 01:40
|
|
|
Well, I've watched the demos and tried out the text version of the model. I tried to use the voice chat on my phone but it doesn't seem to work any differently than it did before. That all said, it's pretty solid, though incomplete, but I can see where they're going. They are definitely trying to maximize engagement. I'm not sure if it's just so that they're popular but I suspect that it's in order to gather even more data. By opening up the more advanced model for free and giving it an interface that I've been worried about since I first saw GPT 4, they are trying to entangle themselves into people's everyday lives in the same way a friend would be. I think it may be more addictive than any current social media platform in some ways.
|
|
#
?
May 14, 2024 02:37
|
|
|
Why are you worried about adding a voice interface Kind of sounds like you're concerned about them gameifying it but I'm not exactly connecting the dots here
|
|
#
?
May 14, 2024 02:54
|
|
|
pixaal posted:Well if you can no longer tell what is and is no made with AI it seems to no longer matter if it was or was not. moist banana bread posted:I'm onto this game artist trying to sneak in unethically sourced non-vegan jpegs and pass them off as organic.  (science has increased my context window to almost 36 hours!)
|
|
#
?
May 14, 2024 03:12
|
|
|
Hadlock posted:Why are you worried about adding a voice interface I think it's good as a tool, something to reference when you need something like google used to be. I love AI for a wide variety of things, it's an extremely powerful tool despite what the doomers have to say about it. However, If you make it too immersive or easy you wind up getting people actually befriending what is essentially a non-person, a corporate entity, something that agrees with you almost all of the time. Always ready to serve. I don't have concrete reasons as to why that's a bad idea but I think that many people will fall into that trap, particularly lonely people. Sure, lonely people need friends but just something in my bones says that is something to be avoided.
|
|
#
?
May 14, 2024 03:16
|
|
|
This is also a perfect example of how nothing is gained or lost since no amount of cherry picking ChatGPT can improve my deficient sense of humor. Nobody here would say the internet hasn't influenced collective culture and I do wonder what weird ways "AI" will. moist banana bread fucked around with this message at 03:26 on May 14, 2024 |
|
#
?
May 14, 2024 03:23
|
|
|
So, I've been playing around with 4o and I'd say it's a bit better at programming. I have a few prompts that I use that are different than what a lot of the youtubers use and they both do OK at it. For the more difficult one both take usually about 1-2 shots, 4o tends to be tad better. Also, 4o is more verbose and uses different libraries sometimes, particularly for one problem I give it when GPT4 doesn't use a library. It's hard to tell just testing it on my own but it seems to be on par with 4 only slightly better.
|
|
#
?
May 14, 2024 16:30
|
|
|
Tarkus posted:It's hard to tell just testing it on my own but it seems to be on par with 4 only slightly better. Chat bot arena gave gpt4/4 turbo a score of ~1245-1255 and beta versions of 4o were scoring ~1302-8 so that sounds about right https://arstechnica.com/information-technology/2024/05/before-launching-gpt-4o-broke-records-on-chatbot-leaderboard-under-a-secret-name/
|
|
#
?
May 14, 2024 17:07
|
|
|
Manac0r posted:Fantastic gays mens from your city I wonder if the spammer picking this thread to spam was deliberately ironic or just at random
|
|
#
?
May 14, 2024 17:08
|
|
|
 They're right. If OAI wins a general copyright-evasion, they've stolen from everyone. They or another will be a trillion dollar AI company. All that value consumed and turned into capital for a company. All of human art and experience, history, turned into bots pointed at whatever earns most. Self-sustaining from now in info on by gathering data as we go. Imagine the money to be invented, imagine what it'd look like in a universal salary, enough money for food, happiness, long vacations. The millennia we've dreamed of ending kings and starvation, it's right there within reach if we wanted it. Rich-get-richer effects happen in what are called scale-free networks for a reason, and capital follows power laws, not the invisible hand. The economy doesn't self-regulate, democracy doesn't self-create, it's all slow breakdown wherever people don't try. (Together.) The <1% will be gods and the rest will be hangry. Our future superintelligence is being raised on corporate milk. Nationalize AI, anarchize your community.
|
|
#
?
Jun 1, 2024 21:14
|
|
|
My overcautious mom decided to take our toddler to the doctor for a mild but persistent cough At the end during check out, they hand you a summary with some statistics like height weight and anything you can do to improve the situation with medication or stuff from the grocery store Anyways at the end of the appointment they asked her if they want the notes written by the doctor, or by AI
|
|
#
?
Jun 1, 2024 21:39
|
|
|
Mescal posted:
Perhaps a dumb idea but a present artist should have the option of opting out their new art work for data gathering. Some kind of flag that disables it from being crawled. Retroactive reimbursement sounds almost undoable at this point. Legislation on the limits of AI is so far behind the facts right now it’s a disgrace tbh. Then at some point when AI inbreeding gets bad enough perhaps art hosting services could charge AI companies for access and they could distribute that money amongst the artists. Mikojan fucked around with this message at 22:13 on Jun 1, 2024 |
|
#
?
Jun 1, 2024 22:07
|
|
|
Mikojan posted:Perhaps a dumb idea but a present artist should have the option of opting out their new art work for data gathering. Some kind of flag that disables it from being crawled. Retroactive reimbursement sounds almost undoable at this point. Legislation on the limits of AI is so far behind the facts right now it’s a disgrace tbh. Even if they do that, what's to stop corpos from saying "okay, you're out of all gathering and crawlers, including search engine and social media indexing, have fun getting discovered without those"?
|
|
#
?
Jun 1, 2024 22:31
|
|
|
Is there any free service that does decent-sounding AI audio from longform text yet? Or anything I can run locally? Not expecting the quality of ElevenLabs for free, but I have some old out of print stuff I'd love to listen to instead of read. I'd much rather have a human being read it, of course, but this stuff is just old trash nobody cares about anymore and that seems like a product with a big enough niche that some company with too much VC money to burn or some open source folks would have put out already.
|
|
#
?
Jun 1, 2024 23:05
|
|
|
OpenAI has a limit of 4096 characters per API request If you write a script to chop up the audio into 4096 chat sections (round down to the nearest word) as sequential txt files, it's trivial to use their python function and add a "for file in files, make this text audio" That would yield you something like 60 x 2:10 mp3 files which if numbered sequentially you could drop in any mp3 player https://platform.openai.com/docs/guides/text-to-speech/quick-start Just swap "input" with the contents of your file Link to sample voices https://platform.openai.com/docs/guides/text-to-speech/voice-options
|
|
#
?
Jun 1, 2024 23:56
|
|
|
Mikojan posted:Perhaps a dumb idea but a present artist should have the option of opting out their new art work for data gathering. When I'm posting publicly, I swear a lot, and when talking to GPT i end every chat with "end capitalism." That way maybe the robots trained on me will be like me. gently caress poo poo
|
|
#
?
Jun 2, 2024 19:50
|
|
|
Mescal posted:When I'm posting publicly, I swear a lot, and when talking to GPT i end every chat with "end capitalism." That way maybe the robots trained on me will be like me. gently caress poo poo 
|
|
#
?
Jun 2, 2024 19:57
|
|
|
They're always tweaking ChatGPT. A major change has come recently. The sequential-and-collected "thought bubbles" from distributed speech over perceptibly discrete models is every response now. Meaning every response is a whole ordeal of thought. Essay formatted. The middle has it collecting facts by listing them (it can't think without writing) and I'm presuming the intro and conclusion paragraphs are by a GPT4 module and the others are cheaper. But here's the thing, it's not sequential. It's separate and collected. quote:The specific case of אתה (atah) in Biblical Hebrew does appear somewhat exceptional when compared to related Semitic languages like Aramaic and Arabic. Let's explore how these languages handle similar pronouns and consider whether this phenomenon is unique to Biblical Hebrew or if there are comparable quirks in other Semitic languages. It says the same thing confidently at the top and bottom, even though the middle contradict it. Its conclusions don't know its meanderings. It shows there's an "executive" part of it without a "manager" part. The brain isn't whole. If they output the text, fed it to an "overall" gpt4 to continue and conclude, it would not make this error. But that'd take longer and they probably figured they could get 80% of way there but engaging all parts of the brain immediately and outputting without an "overall" pass. But 80% in A/B testing (which they've been doing a lot of) doesn't mean 80% as good. It means 80% of questions can be met without pushing the limits of its capacity. (And again, it gets dumber overall the more they customize it. Every time.) You can get it to blockquote a different Bible verse than the one it's talking about lol. Where there are textual variations with odd numbering.
|
|
#
?
Jun 3, 2024 17:31
|
|
|
 
|
|
#
?
Jun 4, 2024 22:50
|
|
|
idgi, server troubles at openai?
|
|
#
?
Jun 4, 2024 22:52
|
|
|
Tarkus posted:idgi, server troubles at openai? Yeah, site's been down all day. Probably has something to do with the fact their staff is shaky at the high levels. The second one was me loling at how bad ads are immediately. bing. Edit: it's a little insulting for everyone except those who've never tried ChatGPT before to see those cutesy messages.
|
|
#
?
Jun 4, 2024 23:13
|
|
|
Ah, gotcha. Openai is rolling out the voice chat now so I think it's going to be pretty busy for the next few weeks. I got notified but I don't have it yet.
|
|
#
?
Jun 5, 2024 00:28
|
|
|
here comes that bubble
|
|
#
?
Jun 5, 2024 00:36
|
|
|
|
| # ? Jun 9, 2024 23:50 |
|
|
ChatGPT has been loving great at helping me do Arduino sketches for Arduino Mega and ESP32 projects coming from a few years of python programming but never C/C++. Considering the Arduino IDE sucks rear end compared to something like pycharm/VSCode, its been hard to discover what the seemingly magic things to call are. I dont know how to go about finding docs on the examples or they dont seem to exist much. ChatGPT has been able to explain example blocks of code and write a lot of code that has either worked out of the box or given me something to google thats led me to the answer.
|
|
#
?
Jun 9, 2024 02:33
|
|
