Stefan Reinalter

Founder of Molecular Matters • C++ & low-level programming • Created Live++ (@liveplusplus)

[Stefan Reinalter @molecularmusing]

Dear Squeaks,

Thank you giving us 2 more days for saying our goodbyes.
Thank you for 10 years of your companionship, your affection, and your love. You were such a kind and gentle soul.

Know that we loved you every single step of our way together. You will be dearly missed.

[Stefan Reinalter @molecularmusing]

@Kappische Putting ketchup on any kind of pasta is a crime and you know it.

[Stefan Reinalter @molecularmusing]

@poikilos_ No, we are not moving goal posts. I never said anything about [] (what does that even mean?) and you thought nobody can come up with better classes than what's already in the STL.

If you know your hardware/platform, the STL is not hard to beat.

[Stefan Reinalter @molecularmusing]

C++ has long had references to (const) arrays and you don't need templates for that.

[Тsфdiиg @tsoding]

Interesting approach to pass an array as parameter without decaying to a pointer in C++. And the function also gets to know the size of it at compile time.

[Stefan Reinalter @molecularmusing]

@poikilos_ Two easy ones:
1) Replacement for std::vector that can grow without copying or moving by making use of virtual memory.

2) Replacing unordered_map with a hash map that doesn't suck.

[Stefan Reinalter @molecularmusing]

@poikilos_ Better than in the standard library?
Absolutely easy for several classes.
Let me know if you want examples.

[Stefan Reinalter @molecularmusing]

@alexsho71327477 @OlexGameDev But that doesn't give you any compile time guarantees, which the original post was about.

I can simply pass an array that is shorter than expected, making the code read out-of-bounds.

[Stefan Reinalter @molecularmusing]

@RLeRoux64 Of course, but why would I?

[Stefan Reinalter @molecularmusing]

@OlexGameDev Yes.
I use that a lot for passing around stuff with a known size, eg formatting/log buffers which have a constexpr size.

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos Still want to add one thing though: I much rather collaborate with humans, have back and forths, give feedback and arrive at something all are happy with than telling a robot what to do.

It was always about the collaborative journey for me.

[Stefan Reinalter @molecularmusing]

@SebAaltonen @btn0s @bwwgpro1 @fuwidabos Fully agree with that Sebastian!
I always gave source code on my blog posts a very permissible license because I wanted others to benefit from it.

But we can't just pretend that AI won't output eg code trained on GPL source, which IMO is very problematic.

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos People would be surprised what an agreement looks like if you give companies special licenses, source code access, etc.

I have agreements with the larger part of the games industry and will say: not touching this with a ten foot pole.

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos So using AI for eg code or assets is totally out of the question for me.

People also often seem to overlook what typically goes into license agreements/EULAs and what you have to promise and declare in terms of indemnificatiom.

[Stefan Reinalter @molecularmusing]

@SebAaltonen @btn0s @bwwgpro1 @fuwidabos I'm pretty sure it's not about an algorithmic solution per se, but rather about the source/training data.

At least it's for me, maybe I should not extrapolate to other people.

As I said, it's very clear to me what eg Speedtree is and does and what AI is.

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos See my other answers. On my phone, please give me a minute to type :D

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos The most glaring issue for me personally: where did the training data come from, how was it acquired, how was it sourced?

AI companies already paid large settlements for training on "stolen data" (in terms of copyright/IP) and don't get me started on destroying millions of books

[Stefan Reinalter @molecularmusing]

@btn0s @SebAaltonen @bwwgpro1 @fuwidabos genAI: dubious how companies acquired training data and whether that data was copyrighted. Muddled copyright, ownership, and licensing of the generated output.

It's simply not comparable.

[Stefan Reinalter @molecularmusing]

@SebAaltonen @bwwgpro1 @btn0s @fuwidabos You all know the answer to that. It's not about automation, there's a very clear distinction (at least to me!) between AI models and eg Speedtree.

[Stefan Reinalter @molecularmusing]

@atorstling Enjoy!
Have to wait until Saturday...