Hey Guys

I came across the article and video below earlier and I am curious as to what you guys think about it, I am kinda 50/50 split.

A part of me is like Holy Cow this is amazing and the other part is like WTF!, especially the podcast creation and live interaction.

Is AI going to far?..... As great as AI can be it really is tredding on peoples toes and potentially peoples Livelihoods in all kinds of ways and maybe making some people lazy.

Original Article

Would love to hear other people's opinions on this,

 

I'm finishing a FreePie plugin that will allow me to send midi though my TrackIr device, so I'll have pitch, yaw, roll, x, y and z movements of my head selectable for CC messages, the MIDI send feature was missing from FreePie so I'm adding it. I have no C++ knwolodeage whatsoever. I've been trying mixing FreePie with Vjoy and LoopMIDI and other workarounds in the past but it was a pain in the ass. Thanks to this new tools I can tackle this project that was totally out of my reach just one year ago.

That's how lazy you can become using Diffusion Models.

 

If it was a downloadable bit of software that didn't collect your data and could be used completely offline I could use it.

 

Very interesting indeed so did it tell you things a manual didnt or did it just understand what you were trying to achieve and then tell you how to achieve it?... have you used this for anything else?

 

Last edited: Feb 24, 2025 at 4:57 PM

Thats certainly something that crossed my mind, although I was thinking "its only PDF manuals" so for my initial use case I have nothing to hide as they are just stock PDF user manuals, although I am sure it would collate all the questions I asked it, and probably record and analyse my voice patterns... this is where it kinda gets grey for me, next thing is some AI is using my voice to do something shady

 

For you both:

This project is free and totally offline if you want (you can give the model access to search the internet as well). It's the one I use. Bigger models require more GPU power but there are some little ones that works really good depending on the task.

https://github.com/open-webui/open-webui

https://openwebui.com/

 

Thanks for the links I will take a look.

Any downsides / concerns for you using this?

 

I use it all the time. If the first thing would be to look up any youtube video, I ask ChatGPT instead. I probably use it most for minimum system requirements for whatever software before acquiring it. I almost never use it for my own gear, but I would do it before I would use a pdf manual copy.

 

Will take longer to generate the responses depending on your rig than using the ones online, but no information is taken away from you if that's something you care about.

 

Reading manuals was a non-negligible part of my learning process.

 

ok thanks for the reply I like your use case... I am going to give the PDF manual upload a try later on on a couple of bits of studio equipment and see how it handles it and what it outputs.

 

I have a pretty top spec rig and soon to have an RTX 5090 to replace my 4090 once local supplier has stock so GPU should be ok.

Privacy is certainy something to consider especially ones voice pattern but I cant think of anything personal I would want to use to upload to it apart from just stock manufactorer PDF manuals.

 

Reading these days just makes me sleepy after about 3 pages and I'm talking A5 not A3 lol

 

I use AI summaries to help me decide whether it's worth reading an article, paper, watch a video or listen to a podcast - and more often than not, it's not worth the time. I don't watch YouTube anymore - I read it.

It hasn't been useful for programming / research for me - all of them get stuff so wrong so badly so often that I'd rather just do the tedious part myself if I need to quadruple check everything anyway.

 

With a 5090 you can run a 70billion parameters model like LLama 3.3 faster as hell. With a rig like that you will not need to be subscribed to any online service. OpenWebUI has more options to tailor made the models than ChatGPT or Claude. You can even create agents with 32GB of VRAM (Agents are two or more models that collaborate bewteen them towards a goal). But if you want RAW power for AI, NVIDIA is going to sell a mini specialiced supercomp with 128GB of VRAM for running models at the same price than the 5090.

https://www.nvidia.com/en-eu/project-digits/

 

Curious to know more about the YouTube, what exactly does it do with YT videos that helps you?

 

Once again interesting stuff thank you.

 

Deepseek maybe? You need to check your computer specs...

https://duckduckgo.com/?t=ffab&q=deepseek+install+locally&iax=videos&ia=videos

The only way these days for preventing data collection is to live without a phone or internet...

 

Summaries. Summaries have been AI's killer feature for me.

- Factual article: Condensed to 7 key points, nothing of importance left out. 12 minutes saved.
- Research paper: 2 minute gist: "Oh, I know this already". 2 hours saved.
- Audio production podcast: Skip the waffling, gimme the important bits. 1h30m saved.
- YouTube videos: immeasurable amounts of time saved (clickbait, self-promotion, waffling, stuff you've heard hundreds of times already from bigger YouTubers. 2h talks from developer conferences, the endless stream of YT links friends and family want you to watch, ...)