that could make a for a nice blog / gist

Arrogance / hubris. Some people even if they know it's rigged will believe they know which side the insiders are on

I suspect it means it's LLM generated without it being checked

Maybe to get a real breakthrough we have to make programming languages / tools better suited for LLM strengths not fuss so much about making it write code we like. What we need is correct code not nice looking code.

> programming languages / tools better suited for LLM strengths

The bitter lesson is that the best languages / tools are the ones for which the most quality training data exists, and that's pretty much necessarily the same languages / tools most commonly used by humans.

> Correct code not nice looking code

"Nice looking" is subjective, but simple, clear, readable code is just as important as ever for projects to be long-term successful. Arguably even more so. The aphorism about code being read much more often than it's written applies to LLMs "reading" code as well. They can go over the complexity cliff very fast. Just look at OpenClaw.

>> simple, clear, readable code is just as important as ever for projects to be long-term successful

Is it though? I'm a long-time code purist, but I am beginning to wonder about the assumptions underlying our vocation.

I guess it's hard to tell until we see more long-term AI-generated project, but many of the ones we have so far (OpenClaw and OpenCode for instance) are well-known for their stability issues, and it seems "even more AI" is not about to fix that.

If you can’t validate the code, you can’t tell if it’s correct.

No?

That's literally the thing they suggested to move away from. That is just an issue when using tools designed for us.

Make them write in formal verification languages and we only have to understand the types.

To be clear, I don't think this is a good idea, at least not yet, but we do not have to always understand the code.

Lean might be a step in that direction.

Yes yes

Let it write a black box no human understands. Give the means of production away.

I have never met an uncreative kid, and studies show kids tend to be more open and creative. But I have to admit I haven't met and interacted with that many average kids, so there maybe some that aren't creative, but a majority are.

Sure and all the same, most people just don't have it.

Well was Jobs a "doer"? Did he get his hands dirty on the code? Or did he use his employees how we would like to use LLMs?

> Well was Jobs a "doer"?

Jobs' talent was that he was an incredibly talented salesman.

Salespeople sell things that already exist. If you can envision new things that would sell well, that's a bit more than sales talent

> Salespeople sell things that already exist. If you can envision new things that would sell well, that's a bit more than sales talent

A lot of gadgets that were claimed by Steve Jobs to have been envisioned by Apple (or rather: by him) - as I wrote: Steve Jobs was an exceptional salesman - already existed before, just in a way that had a little bit more rough edges. These did not sell so well, because the companies did not have a marketing department that made people believe that what they sell is the next big thing.

Have you ever heard of Steve Jobs?

That wasn't too hard for him given he was also an incredibly talented market opportunity spotter and product leader.

Why do people write such nonsense?

Jobs envisioned the iPad and iPhone. Did he do the physical work? No. But he created direction.

Everyone around him at that time has commented on this. Are you going to claim they’re all lying?

> Jobs envisioned the iPad and iPhone. [...] Everyone around him at that time has commented on this. Are you going to claim they’re all lying?

I don't claim that they are all lying, but I do claim that quite some people fell for Apple's marketing (as I wrote: "Jobs' talent was that he was an incredibly talented salesman.").

Maybe it helps to view ffmpeg as a DSL and the ugliness is caused by the CLI constraints / conventions.

For what it's worth, LLMs are a great tool for both composing and understanding ffmpeg commands.

And if you want something more verbose / easier to read you can use something like https://github.com/kkroening/ffmpeg-python (with LLMs) as well

It looks like that project is dead, some googling turned up this one which seems active and popular

https://github.com/pyav-org/pyav

They are not comparable, ffmpeg-python just abstracts away the CLI, pyav is a low level binding of the ffmpeg libs.

It may seem "dead" but ultimately it just helps you build CLI commands in a more sane way, the CLI interface to ffmpeg has been consistent for a long time. Only thing that may change is individual filters which you can just give raw to ffmpeg-python.

I remember when I was heavily using it last year I found a fork that seemingly had more sane typing or something but since LLMs last year didn't know about the newer lib but could write decent ffmpeg-python code I stuck with it and it did the job.

And that same information contained in an LLM is a compression of how many terabytes of training data? Maybe in the future there will be models an order of magnitude smaller and still better performing.

What I'm saying is you can't judge the data in the genome by purely counting the bytes of data.

Or bad players might get owned by better ones, conclude the other guy was cheating and the only way to compete is for them to cheat as well.

Sort of like nuclear weapons

This has happened in online chess, with some people admitting to using engines (ie cheating) to "confirm their suspicion that the other guy is cheating".