lagrange77

I'm with you. I'm saying this should be the goal, but i know it's a utopy. A dystopian cyberpunk future is more likely, in my experience with this planet.

This should be the goal.

Technology exists in the first place to reduce the execution of unwanted tasks by humans. Universal basic income can help here, while the economical model of the world adapts to the pace of technological advancements.

I can relate to this kind of confusion. When your brain expects useful information encoded in a detail, BUTT the author just did it for some trivial reason.

I even sing the 'abcdefg' song to infer letter sort order.

Can anyone recommend resources, preferably books, on this whole topic of building applications around LLMs? It feels like running after an accelerating train to hop on.

That's fantastic. My gradient field is pointing towards it.

Thank you again!

Thank you for the effort you put in your educational work, it helped me and others a lot! In fact, i'm training my nanoGPT version right now. :)

> Ultimately my interest in llm.c is to have a nice, clean, minimal, super dependency-light repo in direct C/CUDA implementation, which I find aesthetically pleasing.

Also, it's awesome that you spend your time on your passion.

Any plans on making a video series on llm.c? :D

I have the feeling that both, symbolic methods and Reinforcement Learning, will be play a major role in more intelligent AI systems.

"Don't work on LLM. This is in the hands of large companies, there's nothing you can bring to the table. You should work on next-gen AI systems that lift the limitations of LLMs."

Even if, or especially if the technology is in the hands of large companies, understanding/studying them is important, not futile.

I think drill sergeants use the same technique.

Enthusiastic woman's voice: I'm sorry, i can't do this. *

Let's sing a song together instead! Watching every motion in my foolish lover's game.. Now you!

* because the code for it was written by an LLM years ago and is now subtly broken without anybody noticing.

Yes it feels very cyberpunk.

Me: I can't breathe, am i going to die?

Enthusiastic woman's voice: Yes Jim, that's absolutely correct! You will die of suffocation in approximately 3 minutes 41 seconds. Anything else i can do for you?

Same here. My car mechanic seems to think i'm an imposter or at least an idiot, when i don't immediately understand his car gibberish.

Take a look at this playlist [0] by MathTheBeautiful. It says it's about PDEs, but it starts with ODEs.

I think most differential equations courses are too focused on symbolic solving techniques. To me, understanding a (physical) system is mainly understanding the differential equation (system) itself, not it's solution. MathTheBeautiful really excels at this approach.

Solutions are of course important as well, but i think that's what computers are for.

[0] https://www.youtube.com/playlist?list=PLlXfTHzgMRUK56vbQgzCV...

Right, i was also like "well, that's even more impressive, than the truck driver becoming a coder at 38 from a few days ago.".

I only understand half of it, but it sounds very interesting. I've always wondered, if the principle of stationary action could be of any help with machine learning, e.g. provide an alternative point of view / formulation.

Ok, thanks. But does not providing - the personal data you can not have or process - to the EU market, make your position any better, legally?

How can a model not be up to GDPR standards? Or are you talking about the services that provide those models? Genuinely interested.

I think this is the best way for most apps. But in this particular case, the main part of the UI is the preview animation, which by principle needs to look and behave the same on every platform, anyway.

I think i would go with Electron in that case.

You get process spawning / IPC access via node and the view layer is easy to implement and change. Especially for 'custom views', which you'll likely want for a fancy/sleek UI. The main downside of Electron imo, the large binary filesize, won't be that important to your target group i guess.

Oh, i get it. I misread it as:

> but you've got an entirely different problem that you could be out of the workforce in between 5-10 years

> but you've got an entirely different problem that you could be out of the workforce between 5-10 years

Do you mean because of automation?

From the article:

> It is likely that existing alternatives and new alternatives will see huge user jumps when that happens.

I've heard that often and i'm pessimistic about it in general. Although there might be some hope, because the channels Invidious users might prefer (nerd stuff) are more likely to move to an alternative, than others.

This is the first time i witnessed, that every single instance doesen't work. Did that happen before?

I think this is the official strategy, i've never seen someone using a different one.

Superior interfaces.

I was wondering if it uses something like vLLM[0] or Llama.cpp[1].

Seems to be Llama.cpp via 'Nitro', which was discussed here before [2].

[0] https://github.com/vllm-project/vllm

[1] https://github.com/ggerganov/llama.cpp

[2] https://news.ycombinator.com/item?id=38887531

Awesome, thank you for publishing it. I was planning the same, but didn't find the time yet.

Looks cool, congrats to the launch!

> compile from compilers

What does that mean?