the8472

35 ms·

661.
by the8472 3y ago
I don't think that's true any longer. Windows now defaults to setting FILE_DISPOSITION_POSIX_SEMANTICS https://stackoverflow.com/q/60424732/1362755
662.
by the8472 3y ago
Well, the kernel does create implicit packetization boundary when you attach FDs to a byte-stream... but this is underdocumented and there's an impedance mismatch between byte streams and discrete application-level messages. You can al
663.
by the8472 3y ago
And SOCK_SEQPACKET which greatly simplifies fd-passing
664.
by the8472 3y ago
Autonomous cars alone will be transformative. And LLMs are not in their end-state, we went from GPT-1 to 4 in 5 years over that timespan it gained lots of new capabilities. Extrapolate, don't just look at what they can do today. Atoms
665.
by the8472 3y ago
> 3. Topics is reducible to (implementable using) third-party cookies. Even with 1st-party cookie jar isolation?
666.
by the8472 3y ago
It's still useful when dealing with high-latency links with non-zero loss. Fine, something might outcompete it, but without it the throughput would suck anyway. E.g. if a service runs on a single server (no CDN) and you occasionally ge
667.
by the8472 3y ago
> (Same applies to SO_SNDBUF/wmem...) If you want to limit the amount of excess buffered data you can lower TCP_NOTSENT_LOWAT instead, which caps the amount that is buffered beyond what's needed for the BDP.
668.
by the8472 3y ago
> Is a "brain transplant" from one LLM to another a thing? At least it's a thing for vision models and was also used to train OpenAI Five (the dota AI)[0]. So it probably applies to LLMs too. [0] https://cdn.ope
669.
by the8472 3y ago
"at least seeing something" leads to jira or teams where you open a thing and then its entries slowly pop in and shift as they get loaded. An illusion of a site being there that can't be used because you can neither focus on
670.
by the8472 3y ago
High Explosive Radio Telemetry
671.
by the8472 3y ago
For ICBM test warheads they have a system that replaces the nuclear pit with a sensor package that measures the incoming shockfront, does signal processing and gets out a radio signal before the entire thing is crushed.
672.
by the8472 3y ago
Nvidia did open source their drivers (or some form thereof) and now provides redistributable firmware blobs. Plus firmware authentication has been broken on some older cards. So hopefully the proprietary drivers will lose importance over ti
673.
by the8472 3y ago
At least one part of the problem is the WebRTC design. It requires a middle man (or side channel) for session initiation. You can't host a static website that does WebRTC between peers because you can't just input an IP+Port to co
674.
by the8472 3y ago
compute + input + visual output: can it run DOOM? visual output: can it animate Bad Apple?
675.
by the8472 3y ago
> That entire stack feels a lot like an attack surface to me, especially given stuff like the fact that windows used to render fonts in the kernel layer. Indeed. https://googleprojectzero.github.io/0days-in-the-wild/
676.
by the8472 3y ago
> Linking the stdlib into every binary doesn't help It should be possible to use a shared library for std but it would come with the limitation that the bins must be compiled with exactly the same rustc as std.
677.
by the8472 3y ago
Using SOCK_SEQPACKET would be saner than pipes or SOCK_STREAM.
678.
by the8472 3y ago
There are many reasons. Failing at extrapolating exponentials. Uncertain thresholds for how much compute and data each individual task requires. Moravec's paradox, and relatedly people expecting formalizable/scientific problems to
679.
by the8472 3y ago
It's like if a squirrel started playing chess and instead of "holy shit this squirrel can play chess!" most people responded with "But his elo rating sucks"
680.
by the8472 3y ago
> There should be better architectures. The drumbeat of progress has been quite steady. On log charts. https://files.catbox.moe/w93was.png
681.
by the8472 3y ago
https://addons.mozilla.org/en-US/firefox/addon/temporary-con...
682.
by the8472 3y ago
quite the opposite. I get a lot of the outputs filtered without any NSFW prompting. > We detected that generated image contains nude content. Try changing parameters.
683.
by the8472 3y ago
Mostly a matter of software not being written to make use of the SSD capabilities. You need parallelism or prefetching to keep the IO queues non-empty. If you have a single-threaded workload which interleaves blocking IO with CPU work and t
684.
by the8472 3y ago
Stability of video models has been a big issue. This is progress, not necessarily the end state.
685.
by the8472 3y ago
It's not necessarily about sustained throughput spent only in this routine. It can be small bursts of processing text segments that are then handed off to other parts of the program. Once a program is optimized to the point where no le
686.
by the8472 3y ago
With AVX10 intel will make the instructions available again on all segments. SIMD register width will vary between cores but the instructions will be there.
687.
by the8472 3y ago
The fact that it's 1) multi-modal and 2) we're approaching text exhaustion does not imply 3) that other modalities are exhausted
688.
by the8472 3y ago
> GPT4 was basically trained on all text Well, there's multi-modal training. There's tons of untapped audiovisual data.
689.
by the8472 3y ago
That drives up costs because you have to maintain many gigawatts of backup capacity sitting idle most of the year. It's better to have more flexible solutions that provide value the whole year. That's either baseload (fission, fus
690.
by the8472 3y ago
Well, neutron bombardment from fusion means you get some radioactive isotopes. It's not as bad as spent fuel but still something to avoid. But that doesn't really matter. This is expensive technology you don't want to blow up

More ›