@pikuma Don't take it personally :-) ... indeed (sadly enough) there are many that are teaching nonsense out there, so statistically speaking he's right. I don't think you're one of them... (seeing what your students were able to build).

@levelsio Unfortunately, this is often (but not always) true... as the saying goes, "Those who can, do; those who can't, teach". Teaching itself isn't the issue - teaching nonsense is. Discriminating may be difficult for a newcomer (checking what the teacher did actually build can help).

I just backed Voxile on @kickstarter

@CraigRood @ForrestTheWoods @FFmpeg Training won't be easy for sure, but I think that once you've a trained LLM-like model then inference should be comparable with current costs. The problem with today training data is that there's not even the CONCEPT of truth in it. Simulation/sandboxing + RL provides the truth.

@Tpouhuk2 C++ guys are way too much obsessed with types. Before C++20 (bit_cast) IMO there was no legal way whatsoever to be able to say read 8 bytes from disk and make an IEEE754 double precision value out of them (e.g. memcpy was ok, but only with bytes coming from a double!)

@iluvconchshells @tsoding What about Common Lisp `return-from` :-)

@ForrestTheWoods @FFmpeg I think the real jump forward will be made when custom specific ASM generation RL models will be trained on actual real or simulated hardware (not specs or articles). IMO is the same between feeding an LLM books on chess/go or having a deep model learning by itself the game.

@ryanjfleury @croloris I tried Zig only a little (wrote a tiny lisp interpreter... a toy I often try with new languages)... about ~400 lines of code. So far I'm NOT impressed with the language, the compilation speed, the resulting executable speed. The resulting code is not simpler than C.

@VoxelPrismatic @IroncladDev I don't really like CSS (I think it's just questionable weird syntax with questionable weird semantic)... but is this use of `|` correct? Claude tells me this would be the correct version...

@mitsuhiko Peace at all costs is amazing, unfortunately only works in certain cases. You cannot talk with someone that punches you in the face.

Nice article. However when you have multiple threads reading and writing in the same location there is an underlying logical problem, and that is why things like `-fsanitize=threads` are amazing and important even on platforms where for example byte access is atomic. Adding an atomic operation at low level, may prevent low-level confusion and data corruption, but often this is just traded in for more high-level confusion and data corruption at a logical level. Just as an example if someone asks an employee the total of the orders predicted for next month, and at the same time asks another employee to update the prices according to a new price list, adding atomicity to the access to each single price doesn't solve that you'll get nonsense totals (some orders using old prices and some using new prices). And even if they don't fight for each single price, still in the end you don't know what the result of the prediction is about.

@JordanJMottaV @tsoding Ok... I think I'll do it anyway just to have something to point to when this topic comes out. But really, just as an example, consider a simple thing as std::vector... and check out the source code in the standard library implementation. That alone is a sign something is wrong.

@JordanJMottaV @tsoding With brainfuck I mean brainfuck the language... and I wasn't trolling at all. Really C++ metaprogramming is just A LITTLE better than brainfuck. I'll record a video about why C++ template metaprogramming sucks horribly, and I'd like your comments.

@JordanJMottaV @tsoding Not really in this specific case... unless you want to argue that "good programming" is also subjective and that brainfuck is therefore a reasonable option. It's not something that fits in a tweet, but really C++ got it COMPLETELY wrong. I'll probably write something about it.

@JordanJMottaV @tsoding That's the dangerous part of it. It's a challenge and challenges can be fun. Unfortunately most the complexity in it is not "real problems", but just made up obstacles... If you think that C++ got good metaprogramming, it means you never saw what good metaprogramming looks like.

@mitsuhiko Java is the language with no concept of pointers but with NullPointerException, right?

@JordanJMottaV @tsoding C++ metaprogramming is nonsense. It was introduced by mistake and remained just as a nerd sniping tool.

@stefanosdeme @bernielomax @mmakaronny @icanvardar Any performance discussion is extremely complex with modern hardware. x64 machine code is in reality broken up in other lower-level instructions, enters a scheduler that does multiple things in parallel. All depending on exact CPU model. See for example

@tsoding I'm not 100% sure replacement is not going to happen. They're not there now, but advancement speed is impressive. What really scares me tho is what will happen after... they're good also because they were fed good stuff, but training an LLM on LLM output doesn't work well...

@davepl1968 Compilers are smarter than you think...