bityard

12 ms·

241.
by bityard 6mo ago
100% agree. Unfortunately tribalism is very trendy right now, especially on social media and online communities.
242.
by bityard 6mo ago
As someone who is not autistic, just tends towards very socially awkward, this is what I do as well. Active listening is a skill I developed by accident out of not having much to contribute to most conversations. As time went on, I saw that
243.
by bityard 6mo ago
No, that is not my argument, in fact I don't have any argument whatsoever. It was just a plausible observation that I felt like sharing. There's nothing further to read into it, I don't have a horse in this race.
244.
by bityard 6mo ago
Er, no, I am fully aware that LLMs have always been non-deterministic.
245.
by bityard 6mo ago
My hypothesis is that some of this a perceived quality drop due to "luck of the draw" where it comes to the non-deterministic nature of VM output. A couple weeks ago, I wanted Claude to write a low-stakes personal productivity app
246.
by bityard 6mo ago
Then the algorithm is very broken for me. I post extremely benign and even somewhat boring things for my friends and family, scroll through the Marketplace scams occasionally, literally never watch Reels, and still--to this day--Facebook th
247.
by bityard 6mo ago
I do the same thing, and I rationalize it with the fact that the clutch in my motorcycle is is constantly bathed in oil so it can take the "abuse."
248.
by bityard 6mo ago
Are you using the same agent/harness/whatever for both Claude and Qwen, or something different for each one?
249.
by bityard 6mo ago
This would likely run fine in just 96 GB of VRAM, by my estimation. Well within the ability of an enthusiastic hobbyist with a few thousand dollars of disposable income. Quantizations are already out: https://huggingface.co/
250.
by bityard 6mo ago
The benchmarks in the model card are purported to be measurements of model quality (ability to perform tasks with few errors), not speed. They almost certainly run these benchmarks on their own cloud infrastructure (Alibaba afaik), which is
251.
by bityard 6mo ago
I have a Framework Desktop too and 20-25 t/s is a lot better than I was expecting for such a large dense model. I'll have to try it out tonight. Are you using llama.cpp?
252.
by bityard 6mo ago
There are infinite combinations of CPU/GPU capable of running LLMs locally. What most people do is buy the system they can afford and roughly meets their goals and then ball-park VRAM usage by looking at the model size and quantization
253.
by bityard 6mo ago
Direct link without the social hop-through: https://codeberg.org/hails/wsl9x
254.
by bityard 6mo ago
Link doesn't work, I guess you meant this: https://frame.work/laptop13pro Also, this one has more upvotes: https://news.ycombinator.com/item?id=47852177
255.
by bityard 6mo ago
Oof, yes, kobo
256.
by bityard 6mo ago
Amazing that it's 2026 and people who write articles on tech-focused sites are still surprised that they never actually owned the books on their Kindle. Do they also think they own their steam games, or Gmail address? I'm a fan of
257.
by bityard 6mo ago
I don't care about every mm of width, and don't understand those that do. A phone up to 3/4" fits into any pocket that a 1/4" one does. I had multiple android phones with replaceable batteries and many were no
258.
by bityard 6mo ago
If you just want to run models, most of TFA is taking the scenic route. All you really need is podman, toolbx, and the Strix Halo toolbox images from https://github.com/kyuz0/amd-strix-halo-toolboxes . Then you just dow
259.
by bityard 6mo ago
You don't HAVE to update the bios over wifi, fwupd is perfectly able to do it as well.
260.
by bityard 6mo ago
The system had RAID, so disks could be replaced without taking anything down. We never had a catastrophic failure, but yeah, and hour of lost data wouldn't be the end of the world. For regular maintenance we would just coordinate with
261.
by bityard 6mo ago
NeoGeo! The system I will always remember as having fun fighting games in the arcade with a home console that only kids with "status symbol" money could afford...
262.
by bityard 6mo ago
As someone who has dealt with wacky storage issues/designs, a lot of this "felt" strange to me. Btrfs? Rsync? Then I got to the bottom and saw that they were only handling about 100 GB of data! At that scale, nearly anything
263.
by bityard 6mo ago
The company I work for provides all engineering employees with a Claude subscription. My job isn't writing (much) code, and we have Copilot with MS Office, plus multiple internal AI tools on top of that. So I'm free to do low-stak
264.
by bityard 6mo ago
Oh, we have a TON of placenames in Michigan that are pronounced much differently than they are spelled. A lot of these are just Native American and French names/words that most people don't know how to pronounce anyway, but a few
265.
by bityard 6mo ago
> Also, black Americans don't call themselves "African-American" unless they were raised in a white environment. Never have. I'm guessing you either don't remember or weren't alive in the 1990's. It was
266.
by bityard 6mo ago
I grew up in a poor rural part of Michigan. Day to day, you didn't meet many people who lived there by choice so the demographic (and Midwestern accent) was pretty homogenous. In high school, our class had a new kid who moved up from K
267.
by bityard 6mo ago
Model developers release open-weight models for all sorts of reasons, but the most common reason is to share their work with the greater AI research community. Sure, they might allow or even encourage personal and commercial use of the mode
268.
by bityard 6mo ago
It turns out, the tiger can't change its shorts
269.
by bityard 6mo ago
Or just clean it sometimes?
270.
by bityard 6mo ago
Yeah, there is some popular misunderstanding about what IPFS is... a lot of people seem to think its essentially free or subsidized distributed cloud storage. But the more you dig into it, the more you realize it's just a fairly ineffi

More ›