Right 😀 Good luck to him. Real life does not work that way. I have seen the quality of AI generated code. It is worse than the code being generated by sweat shops in India, and that is very bad. AI is great for providing an intellisense feel to Vim, code completion. But it cannot be trusted to write applications.
Besides, why would I want to stop doing something I love to do? Telling me good news, you can now stop writing code is sort of like telling the average American good news, you can now stop watching TV 😉 Good luck with that!
@Unus Nemo If only 😄, I post memes just for fun. Totally agree though — AI code quality is still rough. but for me if it could handle chores like cooking and cleaning and other everyday boring tasks, that would be real progress. kek.😇
lol, give it time. I use only local AI and train the LLMs myself. Though, even then I would not trust code produced by the AI just yet in any case. The quality of the material you train the LLM with will definitely show in the results. My LLMs will produce far better code than most commercial AIs that are trained on anything they can scrape from the internet. Though, I do not focus on training my LLMs to code. I mainly train them in philosophy and psychology and nutrition 😉
@Unus Nemo wow, its cool. I’ve been running models like Llama 3.1 8B, Mistral 7B, and Dolphin-Mistral locally for a bit, but I haven't dived into training or fine-tuning them myself yet. i wonder what kind of hardware it requires to train those?
You will not be able to train a base model with it, but you can train with a RTX 5060ti it will just take a while, and of course you will have to train small models. To train larger models I am afraid you have to spend a lot more. At the least you would need a RTX Pro 6000 with 96gb of ddr 7. Which will set you back around $8500.00 dollars. Though that is what is required for serious training.
You can do a lot with a consumer grade gpu these days, though serious AI gpu for commercial use (above the RTX Pro 6000) start at around $16,000.00 which is a bit steep for a hobby 😉.
I train models for my research. I do not monetize my efforts at this point, and may never. Typically, my trained LLMs will be part of a project I am working on. Such as making NPC more realistic in an MMRPG or to answer questions on brewing 😉. My main goal in life it to enjoy life. So I spend a lot of time researching many aspects of life.
If you have a bit of IT skills then you could download llama.cpp which is not affiliated with Meta by the way.
If you have never used git before then start with these basic steps.
Create a fork of the project, to do this you need to make an account on github if you do not already have one. This is not technically required and if you do not intend to modify the code in any way then you can just clone their project directly. You will not be able to sync your local changes with the project though.
I train models for my research. I do not monetize my efforts at this point, and may never. Typically, my trained LLMs will be part of a project I am working on. Such as making NPC more realistic in an MMRPG or to answer questions on brewing 😉. My main goal in life it to enjoy life. So I spend a lot of time researching many aspects of life.
If you have a bit of IT skills then you could download llama.cpp which is not affiliated with Meta by the way.
If you have never used git before then start with these basic steps.
Create a fork of the project, to do this you need to make an account on github if you do not already have one. This is not technically required and if you do not intend to modify the code in any way then you can just clone their project directly. You will not be able to sync your local changes with the project though.
Clone your fork of the project to your system. You will need git installed of course.
Review the building instructions, if you do not have an adequate GPU then you could compile it using the CPU. As long as you only use models that will fit into your memory of your computer you will be fine. Just CPU is a lot slower than using a GPU though I have done it on many occasions on laptops with poor GPUs.
There is a user friendly wrapper around llama.cpp which is called Ollama If you just want to play with local AI it is a great way to go. You could check both out and see which one fits your needs best. Options are a great part of life 😁
Lately I’ve been using AI locally instead of cloud stuff. No internet needed, no accounts, nothing leaving my machine. Cloud AI is cool, but you share more than you realize. Local just feels safer for me. Using models like dolphin-mistral:7b, llama3.1:8b, mistral:7b.
Yes, I only use local AI. llamma.cpp (not related to the Meta llama) and I have used Ollama in the past (which is just a user friendly wrapper for llama.cpp ) though have mostly migrated to llamma.cpp. I also use ComfyUI, I love playing with stable diffusion. I also actually make content I need with it as well 😉.
Honestly, IRC is the most fun I’ve had online. Simple, fast, and no-nonsense. I’ve been using it for about 4-5
months now, mainly connecting to Libera.Chat.
I personally prefer using WeeChat because I enjoy the terminal-based workflow, but there are plenty of other great clients like Irssi or HexChat depending on your style. It’s a great way to find focused communities.
I love the IRC, I have been using it since the late 1990's. Back then web based group chats were incredibly slow and useless. So Usenet and IRC were the place to be.
@liv @SuperDicq They full o protein bruh, if roasted 1st. Thu are on a Journey of life in fedi squatch> you'll see many racism, anti this anti that but in the end no one matters.
An LLM does no such thing. You are thinking of Corporations. An LLM (Large Language Model) is just an abstract of a neural networks. It learns by the information provided to it. Like human Intelligence, artificial intelligence is not perfect. It depends a lot on the information provided and how well it was trained (taught).
AI is like fire. Fire has burned down cities and it keeps billions of people warm and cooks there food. Neither is intrinsically bad or harmful. It is how it is used that can be bad or harmful. If you are upset at the way some people use AI then that is acceptable. I am as well. Though there is no reason to be upset with AI. Be upset with the Corporations and Governments that misuse AI and your grievances will be better grounded in my opinion.
Unus Nemo
in reply to liv • •@liv
Right 😀 Good luck to him. Real life does not work that way. I have seen the quality of AI generated code. It is worse than the code being generated by sweat shops in India, and that is very bad. AI is great for providing an intellisense feel to Vim, code completion. But it cannot be trusted to write applications.
Besides, why would I want to stop doing something I love to do? Telling me good news, you can now stop writing code is sort of like telling the average American good news, you can now stop watching TV 😉 Good luck with that!
liv likes this.
liv
in reply to Unus Nemo • •Totally agree though — AI code quality is still rough. but for me
if it could handle chores like cooking and cleaning and other everyday boring tasks, that would be real progress. kek.😇
Unus Nemo likes this.
Unus Nemo
in reply to liv • •@liv
lol, give it time. I use only local AI and train the LLMs myself. Though, even then I would not trust code produced by the AI just yet in any case. The quality of the material you train the LLM with will definitely show in the results. My LLMs will produce far better code than most commercial AIs that are trained on anything they can scrape from the internet. Though, I do not focus on training my LLMs to code. I mainly train them in philosophy and psychology and nutrition 😉
liv likes this.
liv
in reply to Unus Nemo • •Unus Nemo likes this.
Unus Nemo
in reply to liv • •@liv
You will not be able to train a base model with it, but you can train with a RTX 5060ti it will just take a while, and of course you will have to train small models. To train larger models I am afraid you have to spend a lot more. At the least you would need a RTX Pro 6000 with 96gb of ddr 7. Which will set you back around $8500.00 dollars. Though that is what is required for serious training.
You can do a lot with a consumer grade gpu these days, though serious AI gpu for commercial use (above the RTX Pro 6000) start at around $16,000.00 which is a bit steep for a hobby 😉.
liv likes this.
liv
in reply to Unus Nemo • •Unus Nemo
in reply to liv • •@liv
I train models for my research. I do not monetize my efforts at this point, and may never. Typically, my trained LLMs will be part of a project I am working on. Such as making NPC more realistic in an MMRPG or to answer questions on brewing 😉. My main goal in life it to enjoy life. So I spend a lot of time researching many aspects of life.
If you have a bit of IT skills then you could download llama.cpp which is not affiliated with Meta by the way.
If you have never used git before then start with these basic steps.
Create a fork of the project, to do this you need to make an account on github if you do not already have one. This is not technically required and if you do not intend to modify the code in any way then you can just clone their project directly. You will not be able to sync your local changes with the project though.
Cl
... Show more...@liv
I train models for my research. I do not monetize my efforts at this point, and may never. Typically, my trained LLMs will be part of a project I am working on. Such as making NPC more realistic in an MMRPG or to answer questions on brewing 😉. My main goal in life it to enjoy life. So I spend a lot of time researching many aspects of life.
If you have a bit of IT skills then you could download llama.cpp which is not affiliated with Meta by the way.
If you have never used git before then start with these basic steps.
Create a fork of the project, to do this you need to make an account on github if you do not already have one. This is not technically required and if you do not intend to modify the code in any way then you can just clone their project directly. You will not be able to sync your local changes with the project though.
Clone your fork of the project to your system. You will need git installed of course.
Review the building instructions, if you do not have an adequate GPU then you could compile it using the CPU. As long as you only use models that will fit into your memory of your computer you will be fine. Just CPU is a lot slower than using a GPU though I have done it on many occasions on laptops with poor GPUs.
There is a user friendly wrapper around llama.cpp which is called Ollama If you just want to play with local AI it is a great way to go. You could check both out and see which one fits your needs best. Options are a great part of life 😁
liv likes this.
liv
in reply to Unus Nemo • •Unus Nemo likes this.