3 ms·

I am, personally... at least until it's all done on-device and functions offline. I have no trust with the siphoning of data we see at large I suspect I'm fair

I am, personally... at least until it's all done on-device and functions offline. I have no trust with the siphoning of data we see at large I suspect I'm fairly alone on this. They'll probably do well without me. Most people that even know about it probably don't mind. I can't even verbalize why I do

solomatov 3y ago
>I am, personally... at least until it's all done on-device and functions offline. You could do it now. Apple computers with a lot of RAM are pretty good at running Llama2.
- bravetraveler 3y ago
  Not really, I'm talking mobile phone and near OpenAI quality. My workstation has allowed me to dabble - I'm familiar, a unified pool of memory does very little for me. The experience with self-hosted stuff leaves a bit to be desired, both in generation speed and content. The software needs work, I'm not saying we won't get there... just that we haven't, yet. With a ridiculously beefy system I can eek out some slow nonsense from the machine. It's neat, and I can do it, I just don't find it very useful
alisonatwork 3y ago
This is basically it for me. I've had a play around with some OpenAI-powered sites and it is neat how much it is capable of, but I feel uncomfortable typing personalized prompts or detailed questions into a system where I know everything I type is going to be harvested. You could argue that by commenting on HN or posting anywhere in the internet everything I type is also going to be harvested (perhaps into the very same models), but that contract was always clear. There is a difference between companies using information I have chosen to share publicly, and companies doing the same with what is presented as a private exchange. But once they can fit that mini GPT into my pocket, and the learning it's doing is truly personalized to my own install... for me that will be a much more appealing product. I guess the technology will get there, eventually.

solomatov 3y ago

>I am, personally... at least until it's all done on-device and functions offline. You could do it now. Apple computers with a lot of RAM are pretty good at running Llama2.

bravetraveler 3y ago
Not really, I'm talking mobile phone and near OpenAI quality. My workstation has allowed me to dabble - I'm familiar, a unified pool of memory does very little for me. The experience with self-hosted stuff leaves a bit to be desired, both in generation speed and content. The software needs work, I'm not saying we won't get there... just that we haven't, yet. With a ridiculously beefy system I can eek out some slow nonsense from the machine. It's neat, and I can do it, I just don't find it very useful

alisonatwork 3y ago

This is basically it for me. I've had a play around with some OpenAI-powered sites and it is neat how much it is capable of, but I feel uncomfortable typing personalized prompts or detailed questions into a system where I know everything I type is going to be harvested. You could argue that by commenting on HN or posting anywhere in the internet everything I type is also going to be harvested (perhaps into the very same models), but that contract was always clear. There is a difference between companies using information I have chosen to share publicly, and companies doing the same with what is presented as a private exchange. But once they can fit that mini GPT into my pocket, and the learning it's doing is truly personalized to my own install... for me that will be a much more appealing product. I guess the technology will get there, eventually.