lambda

12 ms·

151.
by lambda 7mo ago
Sorry, I shouldn't have been that harsh about it; I apologize. Doesn't look like I can delet the comment, though. Anyhow, I'm just a bit bitter about the onslaught of obviously vibe coded projects that very little effort has
152.
by lambda 7mo ago
OK, with MiniMax M2.5 UD-Q3_K_XL (101 GiB), I can't really seem to fit the full context in even at smaller quants. Going up much above 64k tokens, I start to get OOM errors when running Firefox and Zed alongside the model, or just fail
153.
by lambda 7mo ago
The example you gave is just so buggy; why do you think that this is worth sharing with the world yet? The very first question is full of obvious bugs. You have 'find . -name "notes.txt"' selected, it then says 'You
154.
by lambda 7mo ago
Huh? Once it gets to the model, it's all just tokens, and those are just in band signalling. A model just takes in a pile of tokens, and spits out some more, and it doesn't have any kind of "color" for user instructions
155.
by lambda 7mo ago
I've literally just gotten Minimax M2.5 set up, the only test I've done is the "car wash" test that has been popular recently: https://mastodon.world/@knowmadd/116072773118828295 Minimax passed this
156.
by lambda 7mo ago
How do you "properly align" a model to follow your instructions but not the instructions of an attacker that the model can't properly distinguish from your own? The model has no idea if it's you or an attacker saying &qu
157.
by lambda 7mo ago
But we didn't have thousands of people suddenly flying in their planes a few months from their first flight. Now, the risks with OpenClaw are lower, you're not likely to die if something goes wrong, but still real. A lot of folks
158.
by lambda 7mo ago
I was responding to the person I was replying to, who confused LLVM with LLM, and who had brought up the slop term. I was surprised that they didn't think it was slop, because of the obvious tells (even with the fixed diagram formattin
159.
by lambda 7mo ago
Yeah, I have a lot of problems with the idea of handing our ability to write code over to a few big Silicon Valley companies, and also have privacy concerns, environmental concerns, etc, so I've refused to touch any agentic coding unti
160.
by lambda 7mo ago
Why do you think ChatGPT doesn't use a quant? GPT-OSS, which OpenAI released as open weights, uses a 4 bit quant, which is in some ways a sweet spot, it loses a small amount of performance in exchange for a very large reduction in memo
161.
by lambda 7mo ago
I'm running Fedora Silverblue as my host OS, this is the kernel: $ uname -a Linux fedora 6.18.9-200.fc43.x86_64 #1 SMP PREEMPT_DYNAMIC Fri Feb 6 21:43:09 UTC 2026 x86_64 GNU/Linux You also need to set a few kernel comman
162.
by lambda 7mo ago
Yeah, this is an AMD laptop integrated GPU, not a discrete NVIDIA GPU on a desktop. Also, I haven't really done much to try tweaking performance, this is just the first setup I've gotten that works.
163.
by lambda 7mo ago
You don't even need to go this expensive. An AMD Ryzen Strix Halo (AI Max+ 395) machine with 128 GiB of unified RAM will set you back about $2500 these days. I can get about 20 tokens/s on Qwen3 Coder Next at an 8 bit quant, or 17
164.
by lambda 7mo ago
The only ways that I can think of to deal with prompt injection, are to severely limit what an agent can access. * Never give an agent any input that is not trusted * Never give an agent access to anything that would cause a security proble
165.
by lambda 7mo ago
> The project owner strongly emphasize the no LLM dependency, in a world of AI slope this is so refreshing. Huh? This is obvious AI slop from the readme. Look at that "ASCII art" diagram with misaligned "|" at the end
166.
by lambda 7mo ago
This isn't the paper that I was thinking of, but it shows a similar trend to the one I was looking at. In this particular case, even down to 5 bits showed no measurable reduction in performance (actually a slight increase, but that pro
167.
by lambda 7mo ago
I mean, GPT-OSS is delivered as a 4 bit model; and apparently they even trained it at 4 bits. Many train at 16 bits because it provides improved stability for gradient descent, but there are methods that allow even training at smaller quant
168.
by lambda 7mo ago
In general, quantizing down to 6 bits gives no measurable loss in performance. Down to 4 bits gives small measurable loss in performance. It starts dropping faster at 3 bits, and at 1 bit it can fall below the performance of the next smalle
169.
by lambda 7mo ago
It looks to me like this is a hosted version of OpenClaw, so you don't need to figure out how to set it up yourself.
170.
by lambda 8mo ago
> I wrote this comment the same way. The LLM fixed 14 grammar mistakes that I think would distract readers more than any LLM-ish phrasing. I don't think that assumption is correct. As you can see by the discussion we're having
171.
by lambda 8mo ago
One thing to keep in mind is that a lot of non-native English speakers use LLMs to translate to English, or to polish their English prose; they may not realize that it causes the translation to come out in a very LLM-style tone. Not sure if
172.
by lambda 8mo ago
It's GenAI. It does something that's kind of like what you asked it to do, but it will skip some details or add other ones or whatever. Dreary architectural pictures will be more likely to have electrical boxes, poor materials, et
173.
by lambda 8mo ago
Yes, vibe coded. The author has posted several other articles about this whole vibe coding project, like this one: https://campedersen.com/brep-kernel
174.
by lambda 8mo ago
I'm guessing by selling fine-tuning, consulting on hosting, and other services? They also seem to be offering their own inference service with their model, obviously as an open weight model that will be commoditized but I'm sure t
175.
by lambda 8mo ago
A man was just shot in the street by the US government for filming them, while he happened to be carrying a legally owned gun. https://www.pbs.org/newshour/nation/man-shot-and-killed-by-f... Earlier they broke dow
176.
by lambda 8mo ago
> how do you install packages/pull images You pre-build the images with packages installed beforehand, then use those image offline.
177.
by lambda 8mo ago
> although we get the very occasional complaint about wanting a dev environment that works offline It's only occasional because the people who care about dev environments that work offline are most likely to just skip you and move o
178.
by lambda 9mo ago
That was a shitpost, it wasn't harassment. And he apologized to you and handed the domain over to you. And now you've decided, 8 years later, to blow up your relationship with a number of other folks in the industry, over a shitpo
179.
by lambda 9mo ago
I eat meat (beef, pork, poultry, and fish) maybe three or four meals a week, and probably about 6 to 8 oz per meal when I eat it. So on a per day basis, yeah, I probably eat about 3-4 ounces of meat per day. But the source you were quoting
180.
by lambda 9mo ago
Thanks, the repo link works now! And I had been looking at it on mobile, which might have been why I was having an issue with the site; now that I'm looking at the site on a desktop browser it looks much better.

More ›