jlokier

32 ms·

301.
by jlokier 3y ago
NimScript comes to my mind. You get most of the Nim language, which is statically typed and versatile. But NimScript is interpreted immediately, like running a Python script, instead of compiled to an optimised binary like regular Nim.
302.
by jlokier 3y ago
Same here. Firefox Focus has worked excellently as my default phone browser for years. I feel confident clicking randomly on cookie popups, knowing all the cookies will be gone by the time I close the page so they don't matter. When se
303.
by jlokier 3y ago
DRAM that is unused according to the OS doesn't need to use battery, as it doesn't need to be refreshed. If this selective no-refresh feature isn't already implemented, I would expect Apple to be working on implementing it in
304.
by jlokier 3y ago
Yes, there are a few technical reasons. They come down to battery, radio and mobile network overheads. Maintaining an open TCP/IP or even UDP/IP network socket to receive incoming notification messages requires sending a packet b
305.
by jlokier 3y ago
> Almost every major complaint or feature request I can think of from the last decade or so of using Discord has been addressed in some way, A feature request I've seen from many people over years, which Discord haven't impleme
306.
by jlokier 3y ago
> It's quite a complex feature and yet I'm not sure any OS ever used it. They did. Linux used x86 hardware task switching and TR (task register) for years. See "struct tss_struct" and "switch_to" in older k
307.
by jlokier 3y ago
> I was impressed, but I also realized that this meant that it was suboptimal for minimizing temporary registers. However, I couldn't think of some algorithm that would multiply it the way I did when I wrote that assembly, since I
308.
by jlokier 3y ago
Here are some positive a and b pairs. a = 2-√3, b = 1 a = 1, b = 1/√3 a = 1, b = √3 At least one √3 always appears somewhere in a or b (if they are non-zero), so they can't both be (non-zero) rationals. In some ways the √3 beha
309.
by jlokier 3y ago
I agree about training time, but bear in mind LLMs like GPT4 and Mistral also have noisy recall of vastly more written knowledge than any human can read in their lifetime, and this is one of the features people like about LLMs. You can&#x
310.
by jlokier 3y ago
They used to support more architectures. Near the top of their wiki page: Were available : armv5tel armv7l (with vfp and neon FPU instructions) powerpc32 Cell SPE (Sony Playstation 3) sparc32 alpha ia64
311.
by jlokier 3y ago
If your motivation is to be able to run the model on-prem, with parallelism for API service throughput (rather than on a single device), you don't need large memory GPUs or intensive memory swapping. You can architect it as cheaper, lo
312.
by jlokier 3y ago
Sibling commenter tvararu is correct. 2023 Apple Macbook with 128GiB RAM, all available to the GPU. No quantisation required :) Other sibling commenter refulgentis is correct too. The Apple M{1-3} Max chips have up to 400GB/s memory
313.
by jlokier 3y ago
> - 96GB of weights. You won't be able to run this on your home GPU. You can these days, even in a portable device running on battery. 96GB fits comfortably in some laptop GPUs released this year.
314.
by jlokier 3y ago
My daily driver is over 10 years old, from 2013. Still going remarkably well. It runs Teams and Zoom just fine. I have thousands of browser tabs. I often use the common office apps; sometimes LibreOffice and Microsoft Office are open at the
315.
by jlokier 3y ago
That issue can be dealt with using an IOMMU under host control. Those limit where the peripheral can DMA regardless of its firmware.
316.
by jlokier 3y ago
That's not the endgame. Eventually each ad has an embedded AI that must be run to see the content after the ad. That AI evaluates whether there's a human who looks and behaves like you watching the ad. The endgame requires your AI
317.
by jlokier 3y ago
I'll just say, I've had my laptop connecting to my phone on tethering 24x7 for years now, since before the pandemic, on a relatively low cost unlimited data contract. I use cellular for both my home and office internet connection
318.
by jlokier 3y ago
The value of OpenAI's own assets in the for-profit subsidiary, may drop in value due to recent events. Microsoft is a substantial shareholder (49%) in that for-profit subsidiary, so the value of Microsoft's asset has presumably re
319.
by jlokier 3y ago
Perhaps. The bylaws provide details such as who can call a meeting, how to perform written votes without a meeting, etc. but not the fact that the board of directors have legal protection as a collectively responsibility group, shielded fr
320.
by jlokier 3y ago
> Last night, Sam got a text from Ilya asking to talk at noon Friday. Sam joined a Google Meet and the whole board, except Greg, was there. Ilya told Sam he was being fired and that the news was going out very soon. I'm very surpr
321.
by jlokier 3y ago
x87 is different from modern floating-point, including the modern floating-point operations on x86 CPUs (SSE and later instructions) and ARM64. x87 calculates with 80-bit floats ("long double" in C), but the host CPU's FPU ty
322.
by jlokier 3y ago
I'm using a Macbook Pro retina from late 2013 as my daily driver, for work as a software developer and other things. It recently passed 10 years, being used every day including weekends. It's been great. I bought it when two non
323.
by jlokier 3y ago
For what it's worth, I was a researcher at CERN with colleagues from multiple countries, where I turned down an offer to do a PhD in particle physics :-) (A foolish career decision in retrospect.) I think they are saying that if you
324.
by jlokier 3y ago
Apple only added hardware AV1 decoders to their new laptops less than a month ago, and to their new phones two months ago. The M2 laptops released in January don't have it. The Apple TV doesn't have it yet, but unlike many set t
325.
by jlokier 3y ago
You can run efficient VM-like distro containers inside VMs (eg. LXD), and if necessary VMs inside VMs. So you only need one rented VM at the top, if your performance needs are low. It probably doesn't make sense economically, due to t
326.
by jlokier 3y ago
It sounds reasonable to me. Around 1994/5 I was doing perspective-correct texture mapping on a Sun SPARC (with real-time dithering too), an Atari Jaguar and a 486 DX4/100. 16k/sec polys on the DX4 (which is only ~500 at 30fp
327.
by jlokier 3y ago
Indeed it's possible, though judging by the release dates, it's not trivial to do right: - Apple implemented iMessage E2EE sync across devices back in 2011. But be careful not to save your chat keys in iCloud backups (local backup
328.
by jlokier 3y ago
I wouldn't say it doesn't offer E2EE by default. It offers private chats, which are E2EE, right there and almost as easy to initiate as the regular chats. I used Telegram in E2EE mode with someone initially, but later we decided
329.
by jlokier 3y ago
I find it even more stunning that there were two cops in the police van, and they both had sex with the handcuffed girl while on duty, confirmed by sperm DNA samples in the girl afterwards. Neither cop got any serious punishment, and thi
330.
by jlokier 3y ago
Miniature Lisps are similar to Forth in this way. Having worked with both I'm inclined to favour Lisps for providing cleaner structure and abstractions for code and data to build up from, with low overhead despite the superfical diffe

More ›