It certainly is fast, but I'm curious if LLMs ever will figure out how bitshifts work..
e.g. from the playground: `static const uint64_t MERSENNE_PRIME = (1ULL << 127) - 1;` which it insists is the correct way to store a 128-bit integer in followup questions.
Neat idea, the 'search' feature is a bit odd though if you don't know which instruction you are looking for. e.g. searching for 'SHA' shows the autocomplete for platforms not selected and then 0 results due to the filters (they haven't been added for SSE/AVX yet), but searching for 'hash' gets you 100 results like '_mm256_castsi256_ph' which has nothing to do with the search.
It's funny that one of their main arguments is that "verified" users used to mean trustworthy information. Like no one remembers Twitter before Musk where you'd get verified by having a blog with the right politics, wrote an article for an outlet the staff liked, was friends with staff, or just straight up paid one to give you the badge..
While not 'evidence' per se, it's a common trend with monopolies (e.g. Xerox back in the day) to shift engineering/R&D budgets to marketing for a better ROI.
IIRC There was some followup hikers that explored and found a cave/entrance near a military area which sort of looked like the described one, only that it had been caved in (which they couldn't judge if natural or man-made). Not sure about the 'validity' or if they just wanted to contribute to the story.
Most laptops these days have a pretty sizable GPU on the same chip. IIRC Triton makes proper use of the Intel graphics while AMDs equivalents work well with OpenCL out of the box. Apple's M1-3 architecture saw some major speedups on llama.cpp etc. as well. Worth noting is that some may need special drivers, my Xeons from 2010 has support for executing OpenCL but needed extra drivers; no comment on modern processors.
Tl;dr: Stenography just helps with obfuscation as long as no-one is looking for it. The second someone looks it'll fail, and if it becomes a common practice; everyone will look for it.
The theoretical idea is that the information could be hidden in a random location, e.g. extracting every Nth bit, such that the information would be indistinguishable from random noise. In practice, both parties need to agree on the location, so it has the same flaw as 'perfect encryption' methods like a one-time-pad, you need to transfer information IRL.
One could make a case for a group/app using a custom scheme to add a layer of security until someone infiltrates the group / RE the app, but you'd get the same security by just encrypting everything with the same password.
Because he's in the lucky position of not needing it. They already split their revenue with creators, accepting such a bribe would not benefit him. So awful PR for no real benefit.
Reminds me of finding an old laptop in the attic (IBM ThinkPad 760LD). 90MHz Pentium, 8MB of RAM, 810MB of HDD, with Win98 installed. I could boot it and start Excel 4 times in the time it took for my workstation to even get to the desktop. Really nice and snappy usage experience.
Sidenote, look up the ThinkPad 760 series, it had some really cool features. Raise-able keyboard so you could have it at an angle, physical sliders for sound and backlight, easy access to the internals and hot-swappable battery.
The SponsorBlock addon lets you whitelist channels you want to support. Also has options to skip self-promotion (including like and subscribe reminders) and quiet parts of videos.