KeyJ

@simontatham@hachyderm.io To be fair, U+221B does look a bit like a pipe with some smoke, so it's at least related, sort-of.

@tinker@infosec.exchange I'm not a VPN user myself, but from what I heard, geoblocking circumvention seems to be the main use case for commercial VPNs.

@grumpygamer They had to make a few concessions to practicality, but they did a pretty decent job :)

@soph The description says that they are adding that footer only if "AI" contributions have actually been used. If that's true (and I'm reluctantly willing to give them the benefit of doubt here), this is an actual improvement, in that it makes it harder for slopcode to be added unnoticed. If you really want to misrepresent generated code as your own, there's still a million ways to bypass this of course, but it's a step in the right direction - or am I missing something?

@chpietsch @lkundrak @schratze The tram was definitely made in ČSSR: https://en.wikipedia.org/wiki/Tatra_T4

@eschaton@mastodon.social @whitequark@social.treehouse.systems Exactly. We used this a lot back then when FireWire's 400 Mbit/s was a substantial upgrade over typical consumer-grade Ethernet's 100 Mbit/s. Lovely times!

@hailey I very much did, back in the day, because a 286 was all I had until 1996. (Yes, ninety-six. Not eighty-six.) (In more seriousness though: Working with DPMI was only slightly less obnoxious than working with 16-bit real-mode code. It was only after switching to "modern" Win32 or Unix-like systems that this segmentation nonsense finally went away.)

@gsuberland@chaos.social That's fake news. Everyone knows that it's just one specific city in Germany which doesn't exist.

@simontatham Convert to both, and print the one with the lowest entropy?

@rygorous Oh, absolutely. Doing a 16x anisotropically-filtered lookup in an ASTC-6x5-compressed cubemap texture with sRGB-to-linear conversion is nothing anyone's ever wanting to do in software. (Except probably as a "fun" exercise, or as some reference implementation. But not in realtime.)

@rygorous Yes, that's what I meant: The individual parts aren't necessarily rocket science, but it's a lot of busywork to do just to sample a frickin' texel (or, more likely, a bunch of texels, in SIMD fashion), with several special-case scenarios that need to be handled as well.

@rygorous "Every FG had a copy of the entire texture memory" was also true for the 3Dfx Voodoo 2 (or the rare dual-TMU Voodoo 1's, for that matter). The entire Voodoo Graphics architecture looks like a massively downscaled RealityEngine to be honest, which is of course no coincidence, considering who designed it.

@rygorous @mogwai_poet I'm not so sure about that. The paper mentions a peak of 240 Mpx/s, which is almost exactly what a TNT2 could do. A GeForce2 is roughly 4x as fast, in terms of raw throughput. The only thing that sets the RealityEngine apart is the absolutely unhinged memory size and bandwidth, which could afford them things like always-on MSAA, and the extremely overpowered geometry processing stage.

@rygorous @mogwai_poet Good point. I couldn't find data on whether the TNT2 did single-cycle trilinear. So, yes, if you consider all the bells and whistles the RE did "for free" (using a *very* technical definition of "free", considering its price tag ;), like trilinear and MSAA, you'd probably come closer to a GeForce 2 indeed.

@simontatham I would choose a simpler metric, like number of different digits, or runs of equal digits. But yes, you won't get 12345678 that way, that's true ... but there will always be counterexamples where whatever heuristic you're using is going to fall apart. I think handling values close to powers of ten and two is already a good start and would be a useful feature.

@simontatham My attempt at a 32-bit formatter: def fmt(x): x &= 0xFFFFFFFF r = [str(x)] if x >= 16: r.append(hex(x)) if x >= 1 << 31: x -= 1 << 32 r.append(str(x)) if x <= -16: r.append(hex(x)) return min(r, key=lambda x: (len(set(x.replace("0x","x"))), len(x.strip("-")))) for x in (0, 17, 257, 4097, 9999994, 100002, 0x7FFFFD, 0x55555, -294967296, -1, -10, -100, -0x555, 1<<29): print(fmt(x)) Lots of weird heuristics though, still passes neither 12345678 nor 0xDEADBEEF.

@aeva I fully get and support your statement, and if a decent(!) Firefox-based browser emerges, I'll be happy to switch. Until then, I'll have to make do with a (comparatively heavily modified) Chromium fork. In fact, that's maybe one of the reasons why Firefox is such a distant second to the Chromium family: It seems to be harder to fork and extend, for whatever technical or legal reasons. I can imagine that some of the Chromium forks out there would rather be Firefox forks if they could.

@static@aus.social @NewtonMark@eigenmagic.net So, according to your IT Security and Privacy policies, using a third-party cloud service for internal communication of trade secrets is fine, but using the data for LLM training is where you draw the line? Not gonna defend Slack's shitty move, but that reasoning seems to be a bit arbitrary too. If you're handing over *any* amount of data to a third party, you can consider that data public. Even without LLM training stuff, there *will* be some kind of breach eventually.

@hp@mastodon.tmm.cx @foone@digipres.club You'd need to unplug the network cable a week in advance for this to work reliably ...