randcraw

19 ms·

271.
by randcraw 2y ago
MONOPOLY. If Youtube officially offers no public service anymore, it's time to vigorously prosecute them for this and their insane number of copyright violations.
272.
by randcraw 3y ago
You do wonder where all the electricity will come from once the full complement of AI and electric cars and heat pumps go online. And how it'll get where it needs to go with a grid that hasn't been appreciably upgraded in decades
273.
by randcraw 3y ago
Isn't that factorial, and much more costly than quadratic?
274.
by randcraw 3y ago
BTW, the author of this piece is an amazing writer. The rest of his blog and the blog roll at 3quarksdaily are worth a closer look.
275.
by randcraw 3y ago
IIRC, the goal of that effort is simply to violate the blood brain barrier, which may have therapeutic value only under very tightly managed circumstances since violating it likely will yield at least as many demerits as merits since the BB
276.
by randcraw 3y ago
I wonder how much change is needed to how we use GPUs for this to yield real gains. Can we revise libcuDNN (and other higher math libs) only, or CUDA itself? Or must GPU hardware change too?
277.
by randcraw 3y ago
It's fascinating that error can accumulate through repeated trainings that 1) is undetected by humans and 2) can degrade LLM or diffusion models (or any transformer model?) so completely. This implies that not only do we not understan
278.
by randcraw 3y ago
Females only?
279.
by randcraw 3y ago
The article completely misses the essential mechanism behind the advancement of science, like cures for disease. Advancing science is not about making better use of the info you have. It's about gathering NEW info you didn't hav
280.
by randcraw 3y ago
Yes, I misspelt.
281.
by randcraw 3y ago
Is there a directory of old growth forests in North America somewhere? I'd love to visit a few and visually compare them to more familiar forests assumed to be mature and wild, but actually are undergoing secession given the extensive
282.
by randcraw 3y ago
This reminds me of a visit to the Burgess Shale back around 1982. It just happened to rain during our visit (not unusual in British Columbia, I'm sure), bringing the Shale's myriad fossils into high relief. We were visibly _surr
283.
by randcraw 3y ago
For perspective, applying for a coding job 30 years ago used to involve ONE phone interview that lasted no more than an hour, followed within a week by an in-person visit at the site. In contrast today's insanely laborious interview p
284.
by randcraw 3y ago
As for treatment/healing: [from the article -- BBB is Blood Brain Barrier] "We have three on-going, major follow-up initiatives. First, we are assessing, in a temporal manner, whether symptom resolution in patients who had long CO
285.
by randcraw 3y ago
In short, turbo supports 128k tokens vs non-turbo's 8k. So if your input fits into 4k (leaving 4k for output), there's no difference.
286.
by randcraw 3y ago
A Google search could do that 20 years ago.
287.
by randcraw 3y ago
I've been compelled to RTO, 3 days a week, at a major pharma for about a year now. There is absolutely NO value in doing this, for me. Social interaction is minimal since our teams are internationally distributed, and parking has bec
288.
by randcraw 3y ago
No. The user base for Apple Music on Windows is huge. Apple devices are secondary. Using the AM app is primary, once you've been successfully sucked into the Apple ecosystem, Borg-like.
289.
by randcraw 3y ago
I've lived in several states with annual safety/emissions inspections (VA, PA). Virginia certainly used to check the aim of headlights (30+ years ago), but probably don't any more. I imagine that the wide range of headlight
290.
by randcraw 3y ago
Based on my strongly negative experience with automatic high beams (2019 Honda Ridgeline), I'm very dubious that smart headlights will hit the intended target, much less continue to do so as the vehicle's hardware ages.
291.
by randcraw 3y ago
Primary information gradient pursuit vectors? I love the notion. IMO, far too little effort has gone into the formal analysis of information landscapes and the efficient navigation thereof in natural language. Humans could benefit greatl
292.
by randcraw 3y ago
If a human can do it, a dog can too, since their noses are so much more sensitive than any human's. I can imagine dogs reliably sniffing out many diseases, then this becoming a standard lab assay, or at least a component thereof.
293.
by randcraw 3y ago
The volume of air that passes through the heat pump must be cooled (or warmed) at the same rate as it is dehumidified, unless humidity control can be done independent of the pump. If you oversize the pump, the house is cooled faster than i
294.
by randcraw 3y ago
The "clickbait" headline corresponds to the paper's abstract pretty well, especially its final sentence: "It is a puzzle why humans tend toward unrealistic optimism, as it can lead to excessively risky behavior and a fai
295.
by randcraw 3y ago
Many states (like PA) have abandoned licence tags to save a few pennies. Unless you're a cop, it's impossible to know if any vehicle's registration is up-to-date. And most cops no longer track this stuff visually either. It
296.
by randcraw 3y ago
As ye sew, so shall ye reap.
297.
by randcraw 3y ago
The limits of math-based modeling inevitably will also apply to AI & ML-based systems. The difficulty of establishing a baseline set of conditions on which to math model a non-linear process (i.e. non-convex) surely will be just as pro
298.
by randcraw 3y ago
Nobody prices the cost of discrete products by the number of years of production. That'd be insane.
299.
by randcraw 3y ago
The cost of each product must include a host of costs that fall outside of the simple manufacturing cost of each. These costs are myriad: R&D, engineering, tooling, maintenance program development, documentation, testing, re-designin
300.
by randcraw 3y ago
1000 aircraft for $1.7 trillion, so each plane costs $1.7 billion. That's nuts.

More ›