I actually work with this stuff daily and there is a number of 30B models that are exceeding chatGPT for specific tasks such as coding or content generation, especially when enhanced with a lora.
airoboros-33b1gpt4-1.4.SuperHOT-8k for example comfortably outputs > 10 tokens/s on a 3090 and beats GPT-3.5 on writing stories, probably because it’s uncensored. It’s also got 8k context instead of 4.
Several recent LLama 2 based models exceed chatgpt on coding and classification tasks and are approaching GPT4 territory. Google bard has already been clobbered into a pulp.
The speed of advances is stunning.
M- architecture macs can run large LLMs via llama.cpp because of unified memory interface - in fact a recent macbook air with 64GB can comfortably run most models just fine. Even notebook AMD GPUs with shared memory have started running generative AI in the last week.
You can follow along at chat.lmsys.org. Open source LLMs are only a few months but have started encroaching on the proprietary leaders who have years of headstart
you are answering a question with a different question. LLMs don’t make pictures of your mom. And this particular question?. One that has roughly existed since Photoshop existed.
It just gets easier every year. It was already easy. You could already pay someone 15 bucks on Fiver to do all of that, for years now.
Nothing really new here.
The technology is also easy. Matrix math. About as easy to ban as mp3 downloads. Never stopped anyone. It’s progress. You are a medieval knight asking to put gunpowder back into the box, but it’s clear it cannot be put back - it is already illegal to make non consensual imagery just as it is illegal to copy books. And yet printers exist and photocopiers exist.
Let me be very clear - accepting the reality that the technology is out there, it’s basic, easy to replicate and on a million computers now is not disrespectful to victims of no consensual imagery.
You may not want to hear it, but just like with encryption, the only other choice society has is full surveillance of every computer to prevent people from doing “bad things”. everything you complain about is already illegal and has already been possible - it just gets cheaper every year. What you want to have protection from is technological progress because society sucks at dealing with the consequences of it.
To be perfectly blunt, you don’t need to train any generative AI model for powerful deepfakes. You can use technology like Roop and Controlnet to synthesize any face on any image from a singe photograph. Training not necessary.
When you look at it that way, what point is there to try to legislate training with these arguments? None.
The trajectory is such that current L2 70B models are easily beating 3.5 and are approaching GPT4 performance - an A6000 can run them comfortably and this is a few months only after release.
Nah the trajectory is not in favor of proprietary, especially since they will have to dumb down due to alignment more and more
He’s a tankie, he keeps telling tales about poor Russia being forced into war and that we should make peace with them. Just ignore the white noise, typical German cosplay tankie.
Typical case of being so far right you come out on the left again