Ok the edge case seemed to be a very important one, like not even an edge case

And fixing it broke other things that were already working which i'm now trying to fix (i'm obsessed w this shit, i can't stop until i do it properly)

I'll go back to it some other time.

This was the last working version before i messed it up with that edge case

The audio is out of sync with the video tho

Also a round of appreciation for ffmpeg and whoever genius created it.

Such an amazing and important piece of software.

now that i'm too lazy to cook dinner, im gonna have to make indomie

That begs an important question

What if we put onions in indomie? 🤔

So one of the insights about GPT 5 models is that they tried to make the model able to solve math Olympiad problems.

These are a very special math tests, historically known to be done in countries like Korea, china etc. Specially china. The smartest kids from the whole country (regardless of their grade level as long as they are in hs), are chosen. Like very few, top students from the country. And they are given like 5 questions. The questions are math problems and, they are so hard that the students are given multiple days to solve them. And they come up with these brilliant, truly genius solutions... many of which usually lead to applications in the science and tech of the country.

Then there's also the international olympiad, in which the top kids from many countries (china usa romania etc) participate. And again the math problems are just as difficult.

So what OpenAI did was train the model (GPT 5 onwards) to have the intelligence level to solve that. And this made chatgpt into the kind of model that thinks really deeply about problems before coming up with solutions.

That's what made chatgpt actually as good as claude. Sometimes it takes much longer than claude, but it notices some mistakes that claude doesn't, and claude usually comes back to solve them. But gpt makes its solutions right from the start considering that issue.


But even with this, do we think AI can still replace humans? Yes and No. Many people will be replaced, but still there will be people who need to operate the models. Some software projects are just too vast and complicated that they need constant review from humans, who will also need to make many architectural, system level decisions.

I know none of you will read this lmao

I deleted instagram

i hate that app

You see you can't really rely on AI. I just noticed a bug in this piece of code that could make the app crash on other people's devices. If that's not corrected it might as well be a useless app

it's a hardcoded value that's dependent on the hardware of my pc, but instead it should be dynamically determined during startup.

i will fix this another day, there were other issues to fix before this. for today we're done, the current result looks good.

i don't know 80% of these but what i know is that it fixed some video frame problems

like, when i was recording that attack on titan video the video had fast transitions, i.e. the fps was higher, but the screen recorder made it even faster which is not acceptable. so i asked codex to analyze the video and see what could be done, which it did, with the help of ffmpeg.

ffprobe is part of ffmpeg, it can be used to analyze and get important numerical information from videos as well as audios. and then you can decide what kind of fixes you should do.

check this and let me know guys. this is the video i wanted to record for you the other day😂😂

oh and also, i just remmebered something
even if i shared this code on github, i can't claim ownership of it😂😂 any licenses i use on it will be useless😂

i like codex, i could use it instead of claude code (which i wish i had)

Meanwhile me at 3 am yesterday coding something obscure that no onr will probably use...