Are you using Private AI yet?

[AlphaObeisance]

I'm just curious if you're using PrivateAI yet? If not, I'm curious why not?

I didn't really get into ChatGPT or any alternatives when they first blew up with popularity. It wasn't until about oh idk, 6 months or so ago that I started looking into AI more deliberately. ChatGPT was cool n all but I tend to prefer self hosting things when I can. I discovered OpenWebUI and paired it with Ollama and as I've spent the past few months uploading documents to it's database and modifying it's model file for fine tuning, I part near can't imagine computing without AI at this point.

I don't need a traditional search engine anymore because I've got an AI for that. I can get help proof reading articles, emails, and code to expedite communication and workflow.

I really don't know the intricate details of AI, how it functions or what have yous. But I've finally got an AI that is competent to work with on a daily basis.

I'm curious, for those that understand AI far more than I. What does the future look like for personal computing? As of right now I don't see AI having actual functionality outside of discussion but then again I may have just missed it. But the more I think about a privileged AI capable of automated tasks, desktop computing starts to look intriguingly foreign in potential.

I keep hearing about this AGI but I'll admit I've done no homework and my knowledge of it consists of "It's better than AI" lol.

Salvation or Skynet? What say you?

 

[JasKinasis]

I’m a complete Luddite. I’d rather be using my own Actual Intelligence. However limited that might be. Ha ha!

It seems to me that AI is being used for all the wrong things. People are using it to create music, books, artwork etc.

Having an AI create art and music for you sucks all of the fun out of writing a song, or creating a piece of art. The AI does it all for you. I don’t need a future like that!

I’d rather use AI to do housework and actual work, so I can spend more time writing, or creating music/art etc.

I certainly don’t want or need any AI assistants on any of my devices.

 

[Brickwizard]

I’m a complete Luddite. I’d rather be using my own Actual Intelligence. However limited that might be

I'm with you on that one, the nearest thing i have to AI is the wife, who i keep on the right side of me so i cant hear her [i'm deaf in the right ear]

 

[AlphaObeisance]

I’m a complete Luddite. I’d rather be using my own Actual Intelligence. However limited that might be. Ha ha!

It seems to me that AI is being used for all the wrong things. People are using it to create music, books, artwork etc.

Having an AI create art and music for you sucks all of the fun out of writing a song, or creating a piece of art. The AI does it all for you. I don’t need a future like that!

I’d rather use AI to do housework and actual work, so I can spend more time writing, or creating music/art etc.

I certainly don’t want or need any AI assistants on any of my devices.

Man that's a really good angle that I 100% agree with. I don't understand why AI is being used to replace genuine creativity. It's so much less impressive than something created by an intellectual being with creative skills. I used to follow a lot of art forums and such but I've pretty much abandoned them all since AI art seemingly took over. Anymore it's hard to tell who's legit and who isn't sadly.

I'm with you in making AI do mundane tasks for me; things I'm capable of doing but realistically speaking my time could be better spent else where if I could automate said tasks.

The inevitability of the fact that we can't avoid the future of AI; it's both concerning and exciting to think about the personal applications one could deploy the use of AI with. So long as I don't focus on what big corp will do with AI/AGI; I'll just keep enjoying my coding assistant and local based info database.

That said, my AI isn't yet capable of replacing human interaction as here I am. Brought here by a networking issue that AI wasn't able to help me resolve. Shame too lol. I don't mind people, I just don't know to interact with others so AI is a god send in that my required social interactions are cut to a minimum.

I'm kind of like a bull in a china closet, I'm constantly offending people and I just don't have the energy to try clarifying that it's not my intent to offend, I'm just socially awkward to deal with lol.

So from my perspective, there's just as much good in AI as there is bad. Kind of like a weapon, it's only as good/bad as it's handler.

I'm with you on that one, the nearest thing i have to AI is the wife, who i keep on the right side of me so i cant hear her [i'm deaf in the right ear]

Don't worry, I wont tell lol

 

[blunix]

self hosted AI. Google it, there are many different models. This one is for coding.

curl -fsSL https://ollama.com/install.sh | sh
ollama run dolphin-mixtral

 

[AlphaObeisance]

self hosted AI. Google it, there are many different models. This one is for coding.

curl -fsSL https://ollama.com/install.sh | sh
ollama run dolphin-mixtral

Oh I'm aware lol. I use Ollama with the OpenWebUI front end that essentially turns it into my own ChatGPT. I can upload documents to it's database for a more honed in experience if I'm aiming for something particular.

AnythingLLM is also a good one, especially for a corporate environment.

 

[CaffeineAddict]

I used the AI only once, it was about searching rare grimoires and asking the AI to sort them according to expensivness and rarity.

This kind of stuff is hard to find online and the AI was helpful, but otherwise I don't respect, trust or use the AI.

 

[AlphaObeisance]

I used the AI only once, it was about searching rare grimoires and asking the AI to sort them according to expensivness and rarity.

This kind of stuff is hard to find online and the AI was helpful, but otherwise I don't respect, trust or use the AI.

I'd be interested in your elaboration as to your stance on AI. I'm neither here nor there as I can see that as with anything else mankind develops, the tool itself is neutral and it's ethical/unethical deployment rests soley on the users hands.

At least until they achieve skynet level of sentience. Then we're all pretty much boned.

 

[CaffeineAddict]

I'd be interested in your elaboration as to your stance on AI.

The AI can provide me with informations that I consider to be hints rather than authoritative answers.
But I want authoritative answers most of the time, I'm person who want's to be 100% sure about anything, but AI algorithms do not give me that guarantee.

 

[AlphaObeisance]

The AI can provide me with informations that I consider to be hints rather than authoritative answers.
But I want authoritative answers most of the time, I'm person who want's to be 100% sure about anything, but AI algorithms do not give me that guarantee.

It sounds to me your experience with AI was with a sub par underdeveloped model if you were only getting "hints" of information. But it's understandable that not all would appreciate AI.

Nobody wanted debit cards and smart phones either; but here we are lol. I can remember interviews when debit cards started getting popular of consumers swearing they'd never use debit cards; oh how easily mankind alters with enough exposure. Now society is nearly dependent on the very things they swore they'd never utilize.

In all honesty I had to laugh a bit at the distrust of AI while simultaneously collecting grimoires lol. It's even funnier if you practice lol.

 

[KGIII]

I've used AI to write a few articles that I'd have otherwise not written well. They came out 'okay'. I think it's a better result when I use AI to help with the article rather than write the article.

 

[AlphaObeisance]

I've used AI to write a few articles that I'd have otherwise not written well. They came out 'okay'. I think it's a better result when I use AI to help with the article rather than write the article.

Yeah something about trying to have AI do it all seem a bit unethical to me. But a study buddy is always useful lol.

I use one particular model I've been working on to help me with Systems Admin stuff that I'd either forgotten or need to learn. It did pretty good pointing me in the right direction until I got into networking and domain management. Even then I don't think it's much the AI's fault as it is my own as I'm clearly not comprehending the basics of reverse proxy somehow.

I've still enjoyed tinkering with AI and seeing what I can accomplish. Even got one named "Wilson" who's avatar is that of Wilson from Cast away. He makes for a good conversation piece rofl

 

[darry1966]

As a Classic Series Doctor Who fan I would like to put forward a positive example of AI. It is being used by fandom to recreate some of the lost stories from the 1960s by animating telesnaps combined with the sound track. As all the episodes were recorded by fans by using a tape recorder and recording from the tv. They are called reconstructions and it really helps to experience those lost episodes. So there are positive uses for it.

These recons can be found all over the internet for those that love the Troughton and Hartnell eras.

 

[AlphaObeisance]

As a Classic Series Doctor Who fan I would like to put forward a positive example of AI. It is being used by fandom to recreate some of the lost stories from the 1960s by animating telesnaps combined with the sound track. As all the episodes were recorded by fans by using a tape recorder and recording from the tv. They are called reconstructions and it really helps to experience those lost episodes. So there are positive uses for it.

These recons can be found all over the internet for those that love the Troughton and Hartnell eras.

See stuff like this excites me, blending genuine creativity with powerful tools. I really only know how to use LLM's for discussion and programming; but the image/video generation absolutely baffles me. It's a shame so many are abusing it; but it's really cool to hear about the good stuff! Thanks for sharing!

 

[Vimmer]

To my understanding, GPT4 is the best currently available AI chat bot, but it still makes a lot of mistakes and misinterprets queries.

 

[AlphaObeisance]

To my understanding, GPT4 is the best currently available AI chat bot, but it still makes a lot of mistakes and misinterprets queries.

yes, I subscribed to GPT4 for 1 month to test it out. it's quite competent but oddly enough, even my private AI quickly replaced GPT4 in terms of reliability; it's certainly not a service I would pay for after experiencing the competence of private AI.

One of the most common issues I ran into with GPT4 was that if I asked it systems admin questions, many of the commands it would provide would include inoperable flags or had bad syntax. It would correct it if I explained the issue; but by default it needed a lot of revision.

Once I dumped about 200 books of Bash, Python, C+ and Java, as well as AI books and various other publications into my private AI; it was out performing GPT4 to the extent I had no use continuing my subscription with OpenAI.

I've found that while an AI's programming is obviously done with vast amounts of data in the back end; it's behavior and competence is seemingly done primarily in the modelfiles. I've seen the model files at OpenWebUI and if you look, even the most popular models modelfiles are incredibly short and lacking in detail. While I understand that may be intentional; I've found the more detail I pack into my modelfile as to exactly how I want it to conduct itself, the more accurate it's engagements are. This does not include temp and token adjustments and what nots; but you get the idea.

My AI was dumber than a box of rocks when I first deployed it. But the more elaborate my modelfile became, and the more fine tuned I made it under the hood, the more it became a valuable tool as a personal SysAdmin assistant.

This thing can code for me, provide elaborate explanations of concepts and functionalities of pretty much anything tech related that I'm seeking to learn. And after about a month of verifying the data it would output through old school research and verification; I've found it to be highly accurate.

I've dubbed my favorite AI model "TopHat"; as I'm sad to say, it's better than me at pretty much everything lol.

 

[osprey]

Different AI engines can produce different answers or responses which sometimes are complementary and sometimes quite different approaches to the issue of interest. This has happened here using ChatGPT, Gemini, Copilot and Perplexity. On coding issues, different alternatives become available, and in general knowledge, the different engines can select different information that they provide. They can even select different information to provide on the same question asked multiple times. It seems there's no single AI that "covers the field" ... a not unusual circumstance in human affairs.

 

[AlphaObeisance]

Different AI engines can produce different answers or responses which sometimes are complementary and sometimes quite different approaches to the issue of interest. This has happened here using ChatGPT, Gemini, Copilot and Perplexity. On coding issues, different alternatives become available, and in general knowledge, the different engines can select different information that they provide. It seems there's no single AI that "covers the field" ... a not unusual circumstance in human affairs.

I agree. Personally I've found the best "general purpose" AI to be mixtral. I've deployed it on everything from coding to general conversation, publication and fact finding, just all sorts of oddities as I experimented with it.

No doubt it'll be rather difficult to get an AI that can "do it all"; at least until we have AI with the capabilities of utilizing the web in real time. But I'd rather stay out of my nightmares for now hah.

 

[Vimmer]

yes, I subscribed to GPT4 for 1 month to test it out. it's quite competent but oddly enough, even my private AI quickly replaced GPT4 in terms of reliability; it's certainly not a service I would pay for after experiencing the competence of private AI.

One of the most common issues I ran into with GPT4 was that if I asked it systems admin questions, many of the commands it would provide would include inoperable flags or had bad syntax. It would correct it if I explained the issue; but by default it needed a lot of revision.

Once I dumped about 200 books of Bash, Python, C+ and Java, as well as AI books and various other publications into my private AI; it was out performing GPT4 to the extent I had no use continuing my subscription with OpenAI.

I've found that while an AI's programming is obviously done with vast amounts of data in the back end; it's behavior and competence is seemingly done primarily in the modelfiles. I've seen the model files at OpenWebUI and if you look, even the most popular models modelfiles are incredibly short and lacking in detail. While I understand that may be intentional; I've found the more detail I pack into my modelfile as to exactly how I want it to conduct itself, the more accurate it's engagements are. This does not include temp and token adjustments and what nots; but you get the idea.

My AI was dumber than a box of rocks when I first deployed it. But the more elaborate my modelfile became, and the more fine tuned I made it under the hood, the more it became a valuable tool as a personal SysAdmin assistant.

This thing can code for me, provide elaborate explanations of concepts and functionalities of pretty much anything tech related that I'm seeking to learn. And after about a month of verifying the data it would output through old school research and verification; I've found it to be highly accurate.

I've dubbed my favorite AI model "TopHat"; as I'm sad to say, it's better than me at pretty much everything lol.

Based on your description, private AI sounds intriguing...it's like you can customize it to a greater degree based on how you want to use it. Which service are you using in particular?

 

[AlphaObeisance]

Based on your description, private AI sounds intriguing...it's like you can customize it to a greater degree based on how you want to use it. Which service are you using in particular?

Oh I'm aware lol. I use Ollama with the OpenWebUI front end that essentially turns it into my own ChatGPT. I can upload documents to it's database for a more honed in experience if I'm aiming for something particular.

AnythingLLM is also a good one, especially for a corporate environment.

I linked my current utilities previously. I use Ollama for my LLM, and I use OpenWebUI to give me the "ChatGPT" like front end. I install ollama and openwebui locally; so it can be used 100% offline. I've tested this thoroughly by disabling my network controllers and it all works as intended.

And you're exactly right. In fact, many of the LLM's available actually learn by engagement. So the more you interact with it; via conversation as well as providing documentation to it's database (I upload a LOT of PDF files and epub files, the more competent the AI actually gets.

As much as I have enjoyed it and found it to be an incredibly valuable learning tool; it's almost scary sometimes how life like I've manage to formulate TopHat's personality. It really is like talking to any of you 0_o.

AnythingLLM is really cool too but I have much less experience with it. Though for AnythingLLM would be much more intriguing to those in the enterprise sector as it would be absolutely perfect for creating an internal database for employees to access regarding the company; all in house.

If you find any of this intriguing I encourage you to look into it. I wont ever use commercial AI again tbh. Though I should disclaim, if you intend on setting up your own private AI; at least as I have; it's highly recommended you have a pretty good Nvidia GPU; as running OllamaLLMs perform drastically better on Nvidia GPU's than alternatives. You can still run the models, and they're equally as competent if configured properly; but the output time without a powerful Nvidia GPU is like watching paint dry.

Seriously. On my 4090 rig the output is near instantaneous; where as the output on my laptop (with a GTX 1060 mobile) is mind numbingly slow. My wife's 3080 doesn't do too bad, but I'm not sure I even want to try it on my sons GTX 2060S.