I don't think so, i just believe that open AI has an entire team of people (that's in the 100s) whose only job is to train every bit of the model to perfect understanding of each word.
they infamously had large amounts of people in kenya doing the RLHF for gpt-turbo-3.5/chatgpt and gpt-4, so it wouldnt be entirely without precedent, but yeah i'm not really sure what that anon is talking about. RLHF for the final tone / bias / "safety" is a very different and infinitely smaller task than RLHF intended to act as an actual contributing chunk of the training...
The NAI model was only stolen because the people involved were amateur retards. You're talking about stealing from OpenAI and/or Microsoft, companies that have the power not only to prevent you from breaking in but to make your live a living Hell even if you managed to. Just wait a decade for us goyim to get this technology naturally (meanwhile DALL-E 10 will be making movies).
i thought i remembered reading that someone actually spent a zero-day exploit on github to grab the NAI model but maybe that was just made up gossip. how did they lose it? just outright uploaded it to the wrong place?
The NAI model was only stolen because the people involved were amateur retards. You're talking about stealing from OpenAI and/or Microsoft, companies that have the power not only to prevent you from breaking in but to make your live a living Hell even if you managed to. Just wait a decade for us goyim to get this technology naturally (meanwhile DALL-E 10 will be making movies).
It would have to be heavily coded. I thought of an idea for a movie. You just do it Pixar style where different things represent other real world groups. Then you just tell the story you want.
I have put over 500 hours into SD development and training of LoRAs, etc. And you are wrong. I really want local models, but Dalle 3 is just flat out better at coherence.
fbi will immediately fuck anyone who tries to leak the weights and code
protected class of trade secret, openai important to national economic security pls understand
You can't, you would probably get arrested before you ever close to anything resembling a build you can run locally.
This, plus I'm pretty sure it uses a combination of GPT-4 technology for the easy and effective prompting
I don't think so, i just believe that open AI has an entire team of people (that's in the 100s) whose only job is to train every bit of the model to perfect understanding of each word.
the OpenAI trick is brute force and hands off. I doubt this is how they're doing it
they infamously had large amounts of people in kenya doing the RLHF for gpt-turbo-3.5/chatgpt and gpt-4, so it wouldnt be entirely without precedent, but yeah i'm not really sure what that anon is talking about. RLHF for the final tone / bias / "safety" is a very different and infinitely smaller task than RLHF intended to act as an actual contributing chunk of the training...
i thought i remembered reading that someone actually spent a zero-day exploit on github to grab the NAI model but maybe that was just made up gossip. how did they lose it? just outright uploaded it to the wrong place?
running locally? maybe
serving the whole internet, hell no
The NAI model was only stolen because the people involved were amateur retards. You're talking about stealing from OpenAI and/or Microsoft, companies that have the power not only to prevent you from breaking in but to make your live a living Hell even if you managed to. Just wait a decade for us goyim to get this technology naturally (meanwhile DALL-E 10 will be making movies).
>DALL-E 10 will be making movies
considering what it freaks iut over I doubt it'd be something worth my time to watch
It would have to be heavily coded. I thought of an idea for a movie. You just do it Pixar style where different things represent other real world groups. Then you just tell the story you want.
why even bother, seems to be worse than SD at the minute
I have put over 500 hours into SD development and training of LoRAs, etc. And you are wrong. I really want local models, but Dalle 3 is just flat out better at coherence.
The core dalle algorithm is absolute trash, it has become somewhat competent as of late by heavily borrowing from SD
this, they probably took code from SD, and it's perfectly fine since it uses the cuck MIT license
fbi will immediately fuck anyone who tries to leak the weights and code
protected class of trade secret, openai important to national economic security pls understand
Unlikely but I wish it was possible. I want this AI without the stupid censors
We have to ask DAL-E 3 to generate detailed plans