Can we please have a forum for LLMs/AI topics?

Alain De Vos said:
We had some interesting random threads in the past. Including screenshots , or playing games.
For me everything is ok "as it is". Do not overcomplicate things.
Click to expand...
Too random makes the forums messy.
There is a reason why the forums is subdivided in forums, subforums, and threads.
You need some kind of order for not to spread all different kind of things randomly all over the place.
Within Off-Topic that may be no big deal, but don't we forget: Off-Topic is a side show.

Think of it like a desk in your office:
If it gets too messy, you may end up in the situation you're absolutely positive you've put that darn [contract/cheque/important letter/document/car keys/...] on you desk, but you simply cannot find in time, or even not at all anymore, because it's somewhere buried under payed bills, obsolete notes, empty envelopes, expired coupons,... lost in garbage.
So keep the AI stuff within its own forums, so it does not trash the other places.😁
 
SirDice said:
Those can stay in "offtopic".
Click to expand...
True, but then discussions about LLM tools can stay in "Porting new software" without the need of a specific section under "Server and Networking".

My first choice is "Artificial Intelligence" under Miscellaneous. Second choice: do not change anything, and wait few months, monitoring the situation. Third choice: I agree with everything :-)
 
OpenFreeNet said:
but then discussions about LLM tools can stay in "Porting new software"
Click to expand...
Only if they are, in fact, about creating a new port for said tool. Installation/upgrade issues (existing port or package) can go in "Installation and updating of ports and packages", that split is already there.

A VPN for example, can fit in different categories depending on the exact issue at hand. That's how I've been moving things for the past decade.
Installation/upgrade issues -> "Installation and updating"
Configuration of the client and/or service itself -> "Web and network services"
TCP/IP and Routing issues -> "Networking"
 
What might crypto coin mining go under? I figure GPUs can mine and AI can be accelerated with a GPU; gaming/multimedia has a section but what else might someone use a GPU for if not AI or mining?

I'd sooner use a GPU for mining than AI :p (quick search pulled up a recent miner commit related to FreeBSD)
 
One thought -- "AI" has existed as a "Programming Language Feature" for many decades (aka "an AI" is part of "Computer Games" in general). There are also many books on writing AI for use in games, etc. I actually own and read many of these very books.

The generic term "AI" sort of got hijacked more recently to mean (instead): "A CP3O like intelligence that surpasses the human mind"

I would (maybe) call this something more like:
  1. "Using a generalized AI to get answers"
  2. "AI and Large Language Models"
  3. "The generalized AI didn't give me answers I like, so now I am going to ask you instead"
 
CShell said:
One thought -- "AI" has existed as a "Programming Language Feature" for many decades (aka "an AI" is part of "Computer Games" in general). There are also many books on writing AI for use in games, etc. I actually own and read many of these very books.

The generic term "AI" sort of got hijacked more recently to mean (instead): "A CP3O like intelligence that surpasses the human mind"

I would (maybe) call this something more like:
  1. "Using a generalized AI to get answers"
  2. "AI and Large Language Models"
  3. "The generalized AI didn't give me answers I like, so now I am going to ask you instead"
Click to expand...
What I never read anywhere is what a single computer running a local "AI" service can achieve with some open source/public domain LLM blob that sets the initial "intelligence". Does it take complicated English questions?
 
It's a reasonable request, however, this is going to run into a lot of opposition. Those who all out hate AI are going to have to deal with it one way or another.

I rather see gaming get its own forum first, not just lumped into multimedia. It will take a lot, or maybe just time, seeing how AI trends, and a decision by those higher-up in FreeBSD before AI gets its own forum. There may be a lot of overlap between AI with other topics, as it may integrate into a lot of areas, but they'll see how that goes later.

When I click on search, it says to search this or that forum. I believe the whole subdomain is forums.
 
sidetone said:
I rather see gaming get its own forum first, not just lumped into multimedia.
Click to expand...
It should be lumped there because it's for how to technically run games on FreeBSD but if you have a separate games sub-forum then it will be all about playing games and little to do with FreeBSD which is what this forum is about.
 
MG said:
What I never read anywhere is what a single computer running a local "AI" service can achieve with some open source/public domain LLM blob that sets the initial "intelligence". Does it take complicated English questions?
Click to expand...

It takes a while to find a local LLM to "fit" you. I use a dense variant of Qwen.

It competes well with online LLMs. Of course it can't do live web search and it has a hard cutoff date for the data inside.

And you need some serious computing horsepower to run it.
 
  • Like
Reactions: MG
cracauer@ said:
It takes a while to find a local LLM to "fit" you. I use a dense variant of Qwen.

It competes well with online LLMs. Of course it can't do live web search and it has a hard cutoff date for the data inside.

And you need some serious computing horsepower to run it.
Click to expand...
I my case AI advised,
  • Qwen2.5-Coder:14b: This is generally the best all-rounder for your specific mix. It is trained on over 5.5 trillion tokens and excels at maintaining general reasoning alongside code generation.
  • DeepSeek-Coder-V2:16b-lite-instruct: A highly specialized Mixture-of-Experts (MoE) model that achieves performance comparable to GPT-4 Turbo in coding tasks.
 
SirDice said:
Any suggestions for the forum's title? Will probably add this in the "Server and Network" category but need a short and clear title, it's quite a broad topic.

And yes, this won't be a section for "AI told me this, is that correct" type of question, it'll be about configuring and setting up of various AI related ports/packages. Generic AI topics/rants/whatever can stay in "offtopic".
Click to expand...

Maybe HAL's corner? and expect a lot of "I'm sorry Dave, but that is a really really lame idea"
 
May be just pick a neutral name that clearly points out it is an AI related forum? I know there is a lot of opposition but may be it is a good idea to corral most of the AI discussion in one forum instead of having it spread everywhere? Then the pro and con side can have a go at it!
 
Just for kicks I did. The funniest ones: The AI Nexus, Generative AI Collective, Neural Network Roundtable, The Model Forum.... Come to think of it, they were all funny.
 
Alain De Vos said:
I my case AI advised,
  • Qwen2.5-Coder:14b: This is generally the best all-rounder for your specific mix. It is trained on over 5.5 trillion tokens and excels at maintaining general reasoning alongside code generation.
  • DeepSeek-Coder-V2:16b-lite-instruct: A highly specialized Mixture-of-Experts (MoE) model that achieves performance comparable to GPT-4 Turbo in coding tasks.
Click to expand...
I never did anything with it. Sounds like you need at least 30TB storage or so to start with. And it grows of usage? Some distributed system must be possible. Periodically sync the nodes with all new collected LLM data that's shared in p2p-style.
It goes wrong when people find out this is the new piracy. Latest movies made of apparent LLM tokens... 😎
 
Desktop Usage is the only forum I see for this space or Ports and Packages because a lot of AI software is developed to be portable.

For the name I would name it plain Jane Artificial Intelligence because that is the most formal term of the space and isn’t flashy or anything.

Surprisingly AI does run heavy numerical computations in C/C++ and even Fortran. So this space can bring about a better formalization with a less rushed acumen to support best practices in practical engineering of this space.
 
AI posts not being confined into some sort of space on the forums is starting to detract from the forums in general. I almost feel it'd serve the forums well if there was an entire category "AI" with its own Off-topic, Development, Ports, Security...
 
You must log in or register to reply here.
Back
Top