ChatGPT vs. My Home AI: Shocking Code Challenge Results

[@marcosoliveira8731]

Best video about the subject so far. Plus the comparison at the end. Really valuable information.

[@Rushil69420]

Those GPUs all together cost more than my car omg

[@Rushil69420]

Nvm I thought those were A100s lmao

[@SB-qm5wg]

The ram is worth more than my car 😆

[@greymalkin]

@TechnicallyUnsure, are these LLM replying with or without internet access? - fan greymalkin

[@TechnicallyUnsure]

Without Internet, once you download the models, you can disable the Internet on the server

[@_symmetry_]

You have an impressive rig.

[@jj-icejoe6642]

Electricity bill ?

[@bluesquadron593]

Yep, where is the power meter? 🤣

[@glenswada]

wow what an incredible setup. Thanks for showing all this. I imagine AI really cannot determine what is a joke or not, so presume its a canned response. But a database full of canned responses to questions that most people would ask is still an incredible thing to have on hand. I imagine something like the Minisforum BD790i SE with lots of ram and a decent gpu would run llama3.3 without costing anything electricity wise. Thanks man.

[@bluesquadron593]

DeepSeek is a Chinese LLM. This may or may not be your concern.

[@ye-1723]

interesting content , btw does these LLMs recieve auto updates or you have to do that manually ?

[@TechnicallyUnsure]

Manually, you can pull newer versions and use the updated models

[@_A_lex_K_]

You don't seem that unsure

[@TienNguyen-ky4dx]

Don't want to buy a ticket so buy a jet plane 😂😂😂

[@TechnicallyUnsure]

More than that, privacy, accessible offline, using with LM Studio, and solving problems and challenges that other public paid LLMs can't solve

[@ghost_of_you_tube]

Bro 😂😂 ... I'm technically unsure if you are batman 😂😂😂

[@anujparekh752]

😂

[@anujparekh752]

He is Batman

[@Ibey_01]

Good! Great video, very useful, I'm thinking about putting together something similar in the future, could you please tell me approximate consumption of a hardware like that?

[@SB-qm5wg]

Nice server rack.

[@RayhanSardar-tj3ji]

Are you batman?

[@TechnicallyUnsure]

Batman? No. But my rig can definitely run Batman Arkham Knight

[@BinaryCloudChaser]

Hardware ? NVIDIA Jetson Orin Nano

[@ghost_of_you_tube]

Questions... 1. An NPU pci it's better this amount GPUs ? (In actual market).😅 2. You Can try image generation models😮. 3. How many minutes it wast to compile, example: linux kernel ? 🤔 5. Can it run minecraft with ultra realistic shaders or any sort of game that can use an multigpu if there is one? 😂😂 6. Are you batman ?😂😂

[@TechnicallyUnsure]

I don't know any NPU card thats as powerful as the cards I have. As for image generation model, I haven't tried, not sure if there are any good ones that are open source. Compiling kernel doesn't use GPU and yes I can probably run any sort of shaders as long as they use GPU.

[@airbeast5671]

Tenstorrent

[@SinisterSpatula]

I'm interested in hosting my own LLM server, but so far I don't use it enough to justify the cost. I'm thinking of making my mac mini m2 into a self hosted llama server. I'd be interested in automations for myself and family, like stuff that would be useful to us 24hrs/365days. Just don't want to drop lots of money on hardware that I know is already behind the curve. There's got to be some consumer inference accelerator cards coming to market in the next few years.

[@dieselphiend]

Why not install something like LM Studio?

[@SinisterSpatula]

🎉a new Technically Unsure!😸 And I'm curious about this topic too. 🍿🍿🍿

[@pavelyankouski4913]

Sphere

[@Jamer1Smith]

not even o1 solved it?

[@TechnicallyUnsure]

Unfortunately no, it kept making changes to the code, fixed one part, corrupted the other, went back and forth with it, but couldn't get it to work properly.

[@Yves_Cools]

@Technically Unsure : your "Hello World" boot message in deepseek cannot be compared to what you did in ChatGPT or other online LLM's since you didn't do the exact same thing in deepseek (you started from another reference point) so your argument about deepseek regarding the boot message machine code is completely useless and means absolutely nothing.

[@TechnicallyUnsure]

No, I did the same with DeepSeek as well, and it gave me the correct code first try.

[@shawnvines2514]

Happy for you, but completely useless for almost everyone else.

[@TechnicallyUnsure]

hmmm... care to elaborate ? Useless how? You can train your own models, fine tune open source models, unfiltered models, not sharing your data, etc. There are many use cases for having such a server, how's this "completely useless"?