wolttam

10 ms·

181.
by wolttam 5mo ago
I have actually, as a parent to a young child My kid was about 3 when he figured out what it means to be a good person, and that's all he seems to want to be now. Maybe it's just where I am, but the majority of parents around me a
182.
by wolttam 5mo ago
I just got into self hosting Deepseek v4 Flash on a single DGX Spark via antirez’s DwarfStar 4 project It feels great to finally have access to something local.
183.
by wolttam 5mo ago
The way I see this going is there will be 10s of thousands of model harness projects out there, because the tools make it so easy to make a harness that suites your workflows exactly the way you like (as someone who made their own harness)
184.
by wolttam 5mo ago
In that case the robot as a whole is the product and the model is just a part of the technology making it possible. That’s the thing; the LLM itself - the chat window - can’t be the whole product for an industry. It’s a technology that you
185.
by wolttam 5mo ago
There's still going to be plenty of use-case and demand for frontier models running across hundreds or thousands of GPUs. It's just not going to be in the current shape - certainly not accessed by the general public for rote busin
186.
by wolttam 5mo ago
> Our internal intuition about right and wrong seems to leave us at an early age Well, mine didn’t. It’s called respect and love your neighbours, treat others as you would like to be treated. It’s pretty simple and feels deeply ingrained
187.
by wolttam 5mo ago
> inspired to write this post by antirez’s recent project DwarfStar 4, which is a version of llama.cpp that’s been stripped down to run only DeepSeek-V4-Flash This is not true, it is its own project. Indebted to llama.cpp, sure, but not
188.
by wolttam 5mo ago
API pricing isn’t cost, we don’t know what cost is.
189.
by wolttam 5mo ago
Nobody here talking about what this represents for demand on these models, if these numbers aren’t made up. One person using 600B tokens in a month. The most I’ve hit is around 500M tokens and I thought that was a huge amount. We’re going t
190.
by wolttam 5mo ago
Their pricing shown is without the discount. > With DeepSeek’s 75% promo applied to current rates, the same run would have cost closer to $0.55, putting it below Kimi K2.6 in absolute cost while scoring 9 points higher. I will be sad w
191.
by wolttam 5mo ago
DwarfStar 4 is DeepSeek 4 (check the repo)
192.
by wolttam 5mo ago
This is just like security; the most secure system is the one that nobody can use. I think the proof-of-work approach that anubis[0] takes is pretty interesting. I love the idea of having to do a small amount of work for the author of the c
193.
by wolttam 5mo ago
In 2023 GPT-4 was allegedly 1.8T parameters. In 2026 we have ~100x smaller models (10-20B) that handily outperform it, and can indeed run on a laptop.
194.
by wolttam 5mo ago
Any measure you put in place can/will be ignored by the actors who never planned to respect your wishes in the first place. That's just how the web works, though.
195.
by wolttam 5mo ago
By my estimation (guess) you won't actually need to spend that much because the models are already getting a point where they don't need to get a whole lot better to be extremely helpful across many domains. And it looks like th
196.
by wolttam 5mo ago
> foobaring, bamblabooing and fghrtawing all the time without telling you what’s going on Oh man, now I have to plug my tool[0]... it doesn't hide anything, but by default tries to provide a pleasant interface (ctrl+o to toggle deta
197.
by wolttam 5mo ago
I dunno, 2.92 seems to bring in some new features and changes that would not typically be brought into a stable release: https://thekelleys.org.uk/dnsmasq/CHANGELOG
198.
by wolttam 5mo ago
Looks like the version in stable is 2.91, which was released within a couple months of trixie. It's not 'ancient' by any stretch. FWIW the fixes referenced here are already fixed in trixie: https://security-tracker
199.
by wolttam 5mo ago
> Also I bet you the headline is a lie. He's already 5k+ LOC into the rust rewrite...
200.
by wolttam 5mo ago
These models understand architecture perfectly well, but they're not trained to care about it when being asked to complex X or Y feature. They're trained to implement the feature in the shortest route possible. So it's not mu
201.
by wolttam 5mo ago
Ok but hating a guy playing with an LLM is a bit extreme. These things are in many ways still just toys (toys that are becoming increasingly useful). People almost certainly send dumber stuff to Claude than this, and just don’t write blog p
202.
by wolttam 5mo ago
It’s relatively easy to get a basic Unix-liked kernel together. Hardware compatibility (and associated testing) is where it gets hard.
203.
by wolttam 5mo ago
These tools let you get a massive codebase functional in 6 days. But, presumably, there's no better language to target than Rust (in terms of safety/performance), and therefore the rest of time can be spent making the birthed-
204.
by wolttam 5mo ago
Certainly companies and governments will have access to better models than the public (in fact, that's already the case with Mythos). The public will still be able to help themselves with models that are behind the frontier.
205.
by wolttam 5mo ago
They get their money from corporate sponsors, because those sponsors benefit greatly from the existence of the Linux ecosystem. They don't need the contributions of individuals to keep going forever and ever.
206.
by wolttam 5mo ago
8 million (~3%) towards the Linux kernel 180 million (~65%) towards ancillary project support, which includes a huge ecosystem of useful technologies around linux Their 'corporate operations' overhead is like 5% of expenses. whoop
207.
by wolttam 5mo ago
I didn't say run in an air-gapped VM... Just as a means to better isolate the workloads I have running (some less trusted than others). Network connectivity and the associated vulnerabilities obviously remain.
208.
by wolttam 5mo ago
I just don't trust the Linux kernel to effectively isolate processes anymore. Don't care if you're using user namespaces, seccomp, etc. There will be a bug. Time for Micro VMs, they're a stronger security boundary (not p
209.
by wolttam 5mo ago
The LLM is mechanistically unable to pick something actually random and outside of its training distribution, so... yep.
210.
by wolttam 5mo ago
Maybe write to the LKML if you have some privy information?

More ›