ashkankiani

I bought my box for $3k last year because I hadn't updated my PC since the first threadripper 1950x, and the Strix Halo is the only x86 with unified memory that's optimized for the kind of CPU intensive stuff I do, and they finally have proper AVX512 too. It was the perfect upgrade for me (except for lack of PCIe 5 & ECC), which coincidentally happened to be perfectly positioned for local inferrence.

and I liked it so much (and saw that the RAM drought was going to last for at least a few years) so I bought a second one to use as a general server for myself and my family.

It can run on one strix halo machine, which is what I have. For everything else, there’s a vibrant community brute forcing more efficient engines and quantizations. There’s a minimum bar right now, so it depends on your hardware.

Even if Deepseek v4.1 is the last open source LLM I have access to, I would be perfectly fine with that. My secret is that I have this relatively unknown technique for solving problems where I use this quantum engine that developed from just atoms over 4.54 billion years into the most sophisticated machine in the known universe. It's a frontier model that every human has access too, actually.

So when I really want to solve a problem, I just think on it. And when I need something to iterate through the things I thought of quickly, I use deepseek.

I'm stupid so explain it in small words for me.

Meta shares responsible for genocide and we're still too fucking scared to go after them for that properly.

I do this with google's bullshit free AI training captchas. If I see a site using those, I never use it again. I've canceled paid services for this before.

Use a proper rate limiting scheme, obviously. You're using access patterns associated with human volumes, so you can only ever scrape at human speeds (except that you get the advantage of constantly saturating the pipe and running 24/7)

That's funny, I got tired of tmux the other day (it got into 100% spin loop from an unsigned int underflow used as a loop condition variable. and I debugged it all by myself like a big boy, without LLMs!) and built something that seems to be identical in architecture in terms of the "supervisor which allocates pty, just stream bytes, client decodes" part. I even use libghostty for the HW accelerated GUI I built. It's the right way to go imo and it's not too difficult to build... hence...

I don't really understand why they would turn this into a company. I become suspicious of products that raise a lot of VC money, implying that they expect a unicorn-ish valuation, where I don't see them being able to justify their valuation via obvious monetization techniques (e.g. selling a classic license for a GUI app where you get updates for some period of time, usually a year, and then you get to keep the software, but no longer get updates, just in case people forgot how things worked before subscriptions).

That usually implies something like selling cloud subscriptions, vendor lock-in, or selling data. I'm happy to be wrong, but it's just an extremely common pattern. And as an aside, I know there's an impulse to monetize everything, but I think we should probably fight that impulse.

Watch Yanis Varoufakis videos on youtube.

That's only if they target Bitcoin and not some other coin which doesn't have an ASIC and has a GPU-friendly hash.

Do you have some personal financial or other incentive to be “aggressively pro AI”?

Convincing the government to regulate their competition while lobbying for laxer requirements for the big guys who have government contracts and get/give kick-backs isn't a far fetched scenario at all considering it's literally happened over and over and over and over again.

I regret not getting addicted to Lisp early. At Berkeley for EECS, the normal CS61A curriculum was based on scheme, but instead we had a guest lecturer from Google who did it on Python. Sure, I got an A+ and I know Python’s internals well, but damn what a miss for me. I much prefer the functional train.

What’s the best way to get started with Common Lisp now for a very very experienced programmer?

Thank you for not forgetting about him. I bring up this insane and tragic persecution and prosecutorial injustice and hypocrisy whenever I can.

I don't understand how people have just moved past all these various obvious flaws in the foundations of our society. It's like finding a bug and saying "we don't need to patch it" and then building on top of it, all while knowing it exists. We all know that something is fundamentally wrong.

You're doing something similar to what I've been prototyping for the automatic metadata enrichment part. So far I've been using deterministic older techniques + manual for my personal data stream, but I'm trying to learn what the state of the art is for efficiently annotating data with LLMs and newer ML.

This seems interesting, but if anyone else is also doing a similar thing, I'd appreciate tips on how to approach it. So far, I think I have some small RAG model in a demo or the very small qwen models. I think fine tuning seems interesting too. Some of the decision models applied in layers (like a binary search on a pre-generated corpus) could be fast enough to help too?

idk too many options, i'm sure this is a solved problem lol