lifthrasiir

6 ms·

61.
by lifthrasiir 4mo ago
The total model size is about 1.2GB (UNet + SDXL VAE included), so probably about ~3GB?
62.
by lifthrasiir 4mo ago
Tried a bit, and while it is very impressive for 0.2B model it would be very hard to convince me that this matches with 10B models. It did work reasonably well with natural images but inpainted regions were visibly smoother than surrounding
63.
by lifthrasiir 4mo ago
I think you are right in the interpretation. In my case though dynamic dispatch was already integrated with bytecode lowering so the error wasn't helpful :(
64.
by lifthrasiir 4mo ago
I had yet another reason to use inline assemblies in WAH [1]. In some compilers (I think it was GCC) intrinsics are imbued with `target("foo")` attributes, which cause forced-inline via `always_inline` attribute to fail somehow. I
65.
by lifthrasiir 4mo ago
I only meant what I said. After all the OP's thesis was that LLMs aren't better than artisanal software engineers, are they? There was no goalpost to move at least in this particular thread. And the solution might be another age
66.
by lifthrasiir 4mo ago
> I have never written a bug that has destroyed my users' hardware, ... Probably whoever (human or agent) originally decided to put TRACE logs into SQLite also thought---or reasoned---so. Maybe the decision was right at that time bu
67.
by lifthrasiir 4mo ago
Intents matter. Compilers can't see through your skull to infer your intents and thus behave very conservatively unless you override that behavior somehow. This inference, alas, also takes (much) time, so compilers have to balance the
68.
by lifthrasiir 4mo ago
The vast majority of that compute is locked in AI accelerators that do the inference. Those hardwares are bad at doing anything other than that---in fact crawlers would need more residential proxies than more computes in that regard.
69.
by lifthrasiir 4mo ago
Of course, but I'm talking about "annoyance". GC type system is especially annoying if you are not writing the full compiler.
70.
by lifthrasiir 4mo ago
I have seriously attempted to write my own WebAssembly 3.0 implementation recently, and while I did finish the whole thing [1] that left me a bitter taste about WasmGC which turned out to be very annoying to implement. In fact, I originally
71.
by lifthrasiir 4mo ago
I think the "obfuscation" is actually two very different acts: the apparent obfuscation, that is concerned with randomly looking output, and the information-theoretic obfuscation, that takes computational effort to undo it. Commer
72.
by lifthrasiir 4mo ago
While this has been downvoted to the death, it is fun to guess how many entries are submitted to each IOCCC. My best guess is around 10^2.5, i.e. 3--400. Rationales: - The number of winning entries and losing entries that get revealed later
73.
by lifthrasiir 4mo ago
Just two months ago I tried to write a short K code with Claude Opus 4.6, only to find that while it had sufficient knowledge about K vocabularies it didn't try to make good use of them. K is, while slightly obscure and obfuscated, a
74.
by lifthrasiir 4mo ago
IOCCC disallows such entries for the obvious reason ;-)
75.
by lifthrasiir 4mo ago
In my experience LLMs were pretty good at deobfuscating many entries (including mine) but very awful at generating any significantly obfuscated code. So obfuscation can be regarded as a truly humane art---at least for now.
76.
by lifthrasiir 4mo ago
Mainly because noisy people are most visible. Both pro-AI and anti-AI (so to speak) crowds have them.
77.
by lifthrasiir 5mo ago
While I agree that Han Unification is not optimal (and fixing them is a welcoming development), it is already too late to reverse it. Even counter-proposals like TRON didn't work at all so far. IVD is the best compromise we can have in
78.
by lifthrasiir 5mo ago
Your argument is absurd because people don't see code---they see glyphs, and using the same code for slightly different glyphs is a non-issue when they are not interchanged. (And when they are interchanged, both would see glyphs "
79.
by lifthrasiir 5mo ago
IVD works, theoretically and practically (recent versions of OpenType have an explicit support for them). It's not their fault that Japanese vendors have been not very quick to adopt them.
80.
by lifthrasiir 5mo ago
For now, follow UCSUR: https://sona.pona.la/wiki/Under-ConScript_Unicode_Registry
81.
by lifthrasiir 5mo ago
Vendors-have-no-capacity-to-handle-more-than-handful-number-of-emojis-per-each-release thing. To elaborate: it should be plain obvious that not every Emoji proposal can be accepted even though all of them are correctly filed, as there would
82.
by lifthrasiir 5mo ago
Han Unification was effectively "fixed" by Ideographic Variation Sequences, so no.
83.
by lifthrasiir 5mo ago
As an example of having not-exactly-a-character as Unicode "characters", it is rather rare that musical symbols are embedded in running texts (which is a primary litmus test for encoding), but musical symbols are typically rendere
84.
by lifthrasiir 5mo ago
Notably though, Linux's requirement (Assisted-by) is different from what Claude Code actually does (Co-Authored-By). I'm not sure, but it might be intentional (to make the signaling explicit).
85.
by lifthrasiir 5mo ago
I wouldn't recommend to paste a completely opaque script though. ;-) For the reference, the uncompressed code is not that big either: async function p(){let e=document.createElement("input");e.type="file",e.onchange
86.
by lifthrasiir 5mo ago
FYI: This has been an integral part of WebAssembly since 2.0. WebAssembly 3.0 also added "relaxed" SIMD instructions which result is allowed to vary under specified constraints at implementation's choice.
87.
by lifthrasiir 5mo ago
Feel free to ignore if you don't feel so.
88.
by lifthrasiir 5mo ago
4.7 turned out to be a disaster in multilingual settings, so I sticked to 4.6 so far. 4.7 seemed to be optimized for (very specific slice of) coding at the expense of everything else.
89.
by lifthrasiir 5mo ago
That might be possible, but Simon Sapin was who tried to specify what exactly are WTF encodings so that should mark sort of milestone.
90.
by lifthrasiir 5mo ago
Good to know that oxipng has surpassed ECT; there is a reason why i put "(still?)" in my original comment. How quick is it? (The main selling point of ECT was that it was way faster than OptiPNG etc. while results being smaller.)

More ›