eigenspace

6 ms·

61.
by eigenspace 2mo ago
I get more value from my smartphone than I used to, and I upgrade it less often. The photos I take with it are genuinely precious to me, and I'm more than happy to shell out 800-1000€ every 3 years so as long as yhe cameras keep gettin
62.
by eigenspace 2mo ago
The article is kinda dumb, and yes this is clearly the area where frontier models having and advantage matters the most, but I'd point out that these smaller open-weight models are performing better than the big Frontier models of just
63.
by eigenspace 2mo ago
LLMs dont do gradient descent to generate tokens. They are trained by gradient descent, but inference doesnt involve it.
64.
by eigenspace 2mo ago
A high quality 1TB NVME SSD costs 164€ on Amazon right now. For a 10 byte symlink, that costs one billionth of a euro. This is like worrying about the money you lose when part of a glass of tap water evaporates.
65.
by eigenspace 2mo ago
You can save a tool call by just making it a symlink
66.
by eigenspace 2mo ago
The parser and lowering passes were in femtolisp, and yes Jeff wrote femtolisp. However, There's now a pure julia implementation of the parser that's on by default, and the julia implementation of lowering is also almost done and
67.
by eigenspace 2mo ago
IMO saying julia is literally a lisp is just not worth the type of discussion it generates. I'd much prefer to say the important part: Julia learned a ton of very important lessons from the lisp family of languages, and has integrated
68.
by eigenspace 2mo ago
Julia is extremely lispy in its semantics and design, it's mostly separated by syntax.
69.
by eigenspace 2mo ago
What sort of projects are you looking at doing?
70.
by eigenspace 2mo ago
Yeah, its not like this js some sort of Open AI company. That'd be ridiculous.
71.
by eigenspace 2mo ago
They mean PCs with Ubuntu installed as the host operating system, as opposed to a PC where Windows 11 is the host OS, and Ubuntu is just a VM inside WSL.
72.
by eigenspace 2mo ago
Tried again and it refused. Maybe I just got lucky bypassing the classifier on the first go.
73.
by eigenspace 2mo ago
Literally just "Dragon Dildo"
74.
by eigenspace 2mo ago
Worked fine for me
75.
by eigenspace 2mo ago
Well, they're cheap compared to Sonnet API costs, but the (heavily subsidized) subscription model for Claude is certainly cheaper.
76.
by eigenspace 2mo ago
I briefly tried some Chinese models with pi.dev. It was fine, I think I mostly preferred the UX of pi as a harness to Claude Code, but my experience with the chinese models was worse than with Claude's models. They just felt significan
77.
by eigenspace 2mo ago
I think if there's going to be advantages to making smaller, more targeted models, those advantages will probably come from targeting specific domains, not from targeting specific languages. I think that if an LLM can't abstract o
78.
Germany Sets New Six-Month Startup Record(gtai.de)
45 points by eigenspace 2mo ago|21 comments
79.
by eigenspace 2mo ago
Very cool stuff, I've long thought there must be some fancy ways to get extra information for better image resolution by taking advantage of the other cameras. That said, I must also say that most of the side-by-side comparison photos
80.
by eigenspace 2mo ago
It turns out that when you make a massive profit motive to sell expensive fancy interventions, health outcomes suffer. You see this to an even more extreme degree amongst the ultra rich and powerful, who are often using cutting edge, experi
81.
by eigenspace 2mo ago
Sure, but that's not what's driving these increases in costs.
82.
by eigenspace 2mo ago
Sure, this is of course true, but there's also a reason that Americans spend way way more on their healthcare than any other developed nation, and it's not that they have better healthcare, or more healthcare options.
83.
by eigenspace 2mo ago
I think one problem with that comparison is that due to how energy intensive these datacentres are, they are very fast depreciating assets, unlike e.g. the telecom infrastructure that was (over-)built during the dot-com bubble, or the railw
84.
by eigenspace 2mo ago
No, I mean on an age-corrected basis, throughout the spectrum of ages. i.e. young americans are spending more on health care than young americans used to, even if it's also true that there are fewer young americans now, and more old am
85.
by eigenspace 2mo ago
Fair enough, though I think it's rather unlikely that they'd be lying about their work on a big model, even if one might be reasonably skeptical if the model will be any good. I think it's rather important for them that they
86.
by eigenspace 2mo ago
It's not just boomers (though they are the biggest part of it). Americans of all ages are spending more on healthcare, and seeking more healthcare interventions than in previous decades.
87.
by eigenspace 2mo ago
Healthcare has consistently been one of the biggest drivers of US GDP and job growth numbers for years now. This is one of the major things that has helped explain why US consumer sentiment has been so out of step with headline numbers like
88.
by eigenspace 2mo ago
Regarding > Mistral seems to have given up on reaching the frontier and focuses instead on niches like moderation, OCR, and robot navigation. That's not actually true. They at least claim they have a new attempt at a large model the
89.
by eigenspace 2mo ago
The very very northern tip of Germany reaches almost as far north as the southernmost tip of the Alaskan panhandle, but most of Germany is more like mid-south Canada on a latitude basis. Northern Germany is much more specialized into wind p
90.
by eigenspace 2mo ago
Right, I just mean I wonder how much of the buildup to this project will be wasted.

More ›