The media in this post is not displayed to visitors. To view it, please log in.
In which the ignorant weasel in charge decides to eliminate actual heroes to prevent comparison, then does it the stupidest way possible as publicly as possible...
Pete Hegseth cuts the pension and strips Senator Mark Kelly of his military rank, because he — a democrat— had the audacity to disagree with the Trump regime.Freedom of speech, ladies and gentlemen.
Freedom of speech.
Blaise reshared this.
Unus Nemo
in reply to Blaise • •@Blaise
I have no issue with AI, the technology, I have been working on it since 1986. The development of LLMs, artificial neural networks has certainly been a game changer. Though I am not any part of the obsession to make huge server farms that power AI Instances to provide services for mobile devices, or any device you want to hook to them. I use only local AI that I work on, or those that collaborate with me work on.
One very disturbing issue for me is that people cannot make a distinction between AI, the technology, and services provided by corporations. There is a huge difference. For instance I do not train my AI on copyrighted material stolen on the internet. I train them on open source, public domain resources and my own content.
I use ComfyUI (generative AI) to make assets, and just to have fun. I use only checkpoints and loras trained on public domain content, and yes, you can check this. I have used LLMs that I have trained to turn
... Show more...@Blaise
I have no issue with AI, the technology, I have been working on it since 1986. The development of LLMs, artificial neural networks has certainly been a game changer. Though I am not any part of the obsession to make huge server farms that power AI Instances to provide services for mobile devices, or any device you want to hook to them. I use only local AI that I work on, or those that collaborate with me work on.
One very disturbing issue for me is that people cannot make a distinction between AI, the technology, and services provided by corporations. There is a huge difference. For instance I do not train my AI on copyrighted material stolen on the internet. I train them on open source, public domain resources and my own content.
I use ComfyUI (generative AI) to make assets, and just to have fun. I use only checkpoints and loras trained on public domain content, and yes, you can check this. I have used LLMs that I have trained to turn
vimon my machine into a full blown IDE (not that it was not already). There are many practical implementations of AI that does and will enhance all of our lives. That does not mean I support corporate server farms trying to inject AI into all aspects of our life, need or want it or not. On the other hand I do not feel that it is fair for people to be judging AI for what corporations do. Should I judge fire for what arsonist do? Should I judge antibiotics by their current abuse in agriculture?It is an old technology that we have been working on for decades. With great success in recent decades. It is not SLOP it is valuable technology when used appropriately. AI should not be held accountable for the way corporation misuse it for monetization. Corporations should be held accountable for their actions. Not the millions of developers that have diligently worked to hone a technology that is quite valuable.
Unus Nemo
Unknown parent • •@plan-A
Yes, except I write my own extension and do not use a plugin manager 😉
Unus Nemo
Unknown parent • •@plan-A
There is nothing wrong with using
vim-plug. It just does not do anything for me as I write most of my own plugins. The main reason to usevim-plugis for keeping plugins updated. Today, that is moot, as even tpope has not updated a vim plugin in over 4 years. I tend to hard fork the plugins I like best and maintain them myself. So a plugin manager is not required.Yes, with low ram your options are rather limited. I find it is best to train your own LLM though that requires a significant GPU. At the very least an RTX 3060i or equivalent. You just have to be careful what open source, or your own source you train the LLM on. As there are a lot of low quality projects. Train with a bunch of low quality code, get low quality suggestions.
Though I mainly use AI for an Intel-sense level of auto completion. Having an AI write actual code is very risky. With anything even slightly complicated you are bound to do more debugging than it is worth. It would be less effort to just write it from scratch yourself.
Unus Nemo
in reply to Blaise • •@Blaise
You can ask the AI how to train it, it knows 😉
Unus Nemo
in reply to Blaise • •@Blaise
I have a server with two Tesla K80s they're are ancient but great for training. The big issue is power, they use 300 watts each. So I have to have a 1300 watt power supply for the rig, they are also intense on heat, so you have to plan for that in your build! You can typically get them for around $80.00 USD each though you have to build for them, and you are going to have to know how to build and maintain a lot of older code that has been abandoned as there is not a lot of support for the Kepler architecture these days.
Other factors of the build are that K80s are GPU accelerators so they do not have video ports. Server mother boards do not have built in Graphics so you will have to have an addition card for that. I use a
ASUS NVIDIA GeForce GT 730 Graphics Card (PCIe 2.0, 2GB GDDR5 Memory, 4X HDMI Portson a PCI x 1 port just to attach a monitor if needed. Though I do not typically leave a monitor attached to my server.In budge
... Show more...@Blaise
I have a server with two Tesla K80s they're are ancient but great for training. The big issue is power, they use 300 watts each. So I have to have a 1300 watt power supply for the rig, they are also intense on heat, so you have to plan for that in your build! You can typically get them for around $80.00 USD each though you have to build for them, and you are going to have to know how to build and maintain a lot of older code that has been abandoned as there is not a lot of support for the Kepler architecture these days.
Other factors of the build are that K80s are GPU accelerators so they do not have video ports. Server mother boards do not have built in Graphics so you will have to have an addition card for that. I use a
ASUS NVIDIA GeForce GT 730 Graphics Card (PCIe 2.0, 2GB GDDR5 Memory, 4X HDMI Portson a PCI x 1 port just to attach a monitor if needed. Though I do not typically leave a monitor attached to my server.In budget build you would likely be better off with P100 They have the Pascal architecture v. Kepler. More recent but still weakly supported. It out performs the K80 using less power as well.