senbrow

Hah, that exact transition was my last project there before I decided I had had enough!

The GPM team was hugely passionate about music and curating a good experience for users, but YT leadership just wanted us to "reuse existing video architecture" to the Nth degree when we merged into the YT org.

After literally years of negotiations you got... what YTM is. Many of the original GPM team members left before the transition was fully underway because they saw the writing on the wall and wanted no part of it. I really wish I had done the same.

As someone who used to work there, Google will never get product releases right in general because of how bureaucratic and heavyweight their launch processes are.

They force the developing team to have a huge number of meetings and email threads that they must steer themselves to check off a ridiculously large list of "must haves" that are usually well outside their domain expertise.

The result is that any non-critical or internally contentious features get cut ruthlessly in order to make the launch date (so that the team can make sure it happens before their next performance review).

It's too hard to get the "approving" teams to work with the actual developers to iron these issues out ahead of time, so they just don't.

Buck passed, product launched.

I'm pretty sure most PS controllers use Bluetooth natively if they're not connected via USB.

Thanks for writing such a detailed reply about this.

Like GP I'm not very good at spotting these patterns yet, so explicit real-world examples go a long way.

Almost everyone is using these controllers wirelessly if I had to hazard a guess.

The USB interface is used for initial pairing and charging, in which case the port location doesn't matter nearly as much.

Total nitpick, but:

Why would you name an option "disable_ai" with a default value of false instead of calling it "enable_ai" with a default value of true?

Are there some mechanical semantics I'm missing here that make this beneficial?

Negative booleans (ie that remove or suppress something when true) are generally a source of confusion and bugs and should be avoided like the plague in my experience.

Did you even look at the content of the post?

There's a prompts.md (link: https://github.com/maciej-trebacz/tower-of-time-game/blob/ma...) that shows what the author did in this project, most of which is providing creative direction and corrective nudges to the AI.

Most people only care about the end result, and vibe coding got the author there much faster and with less effort.

Your comment reads like a carriage driver bemoaning car drivers because they didn't have to feed, groom, harness, and command a team of horses, yet still arrived successfully at their destination.

"Why should I be impressed that you turned a wheel and pushed a pedal?"

Long names are good for short expressions, but they obfuscate complex ones because the identifiers visually crowd out the operators.

This can be especially difficult if the author is trying to map 1:1 to a complex algorithm in a white paper that uses domain-standard mathematical notation.

The alternative is to break the "full formula" into simpler expression chunks, but then naming those partial expression results descriptively can be even more challenging.

To those who want to do this: make sure to swap out all cells in a battery at once to be safe, ideally using new cells that are all from the same manufacturer and same production run.

Mixing old cells with new can lead to a runaway thermal event in the worst case (i.e. unstoppable cancer-causing fire).

Why couldn't you use an LLM to generate source code from a prompt, compile it, then train a new LLM on the same prompt using the compiled output?

It seems no different in kind to me than image or audio generation.

...by a human :)

At that point why not just have LLMs generate bytecode in one shot?

Plenty of training data to go on, I'd imagine.

1000% - I had to be able to find something listenable

You're definitely in a bubble if you think "nobody" considers AI useless.

There are a huge number of non-technical people who see its use as a dangerous outsourcing of human thinking and creativity to a machine that is only capable of producing derivative (and often incorrect) slop.

Why should I appreciate a company's exploitative and extractive experimentation on its customers?

The cost of maximizing "value" for the company to the nth degree degrades the customer experience once it exceeds a certain threshold.

It's greedy tunnel vision that makes the world worse for everyone in the long term.

The ebike analogy is perfect!

You still have to pedal, steer, and balance, but you're much faster overall.

35 year old US resident here - the recommendation algorithm is incredibly good at being very quickly tuned to your interests, so if you learn to take that seriously it's an incredible tool for learning about different crafts / trades / topics from experts who create content.

For example, my feed was full of electrician and HVAC content, philosophy, skiing tutorials, constitutional law, international news, physics, etc.

You have to use the app with intention to arrive at and maintain this state. I actively press the "not interested" button on videos that are mindless or just not in line with the kind of experience I want (rage bait, suggestive content, brand content, etc). That's a skill that many non-TikTok users don't really understand as well I think.

The algorithm is very much like an AI companion in some ways, which can be good or bad depending on the user's maintenance and steering of that relationship.

One strategy that sometimes works for me (ADHD) is the "momentum building" approach.

Rather than start with the hardest tasks first, I'll start with the easiest.

If I'm struggling with the easiest one, then it means I need even easier ones on the list to ramp up.

If I'm REALLY stuck, I can often overcome that starting inertia by doing something physical IMMEDIATELY (pushups, body weight squats) and then trying to see how fast I can start the task I'm stuck on. Something about getting the body moving seems to help.

There is also the trick of doing some introspection about why I'm having trouble starting a task, because often my executive dysfunction is tied to some emotional valence (eg fear of messing up, fear of taking too long because I will get distracted) and addressing each one aloud ("It's okay if I mess it up, I'm going to try it anyway", etc).

I suppose I was imagining tests that are created to validate an internal piece of code when it's first built or refactored, not necessarily in pursuit of a specific observed bug.

Hmm, I guess I don't really consider throwaway assertions in pursuit of debugging "tests" in the typical sense.

I certainly wouldn't advocate for code reorganization for that case, but if there is a property that is important to maintain over time that isn't easily expressed by exercising the public API, it does suggest that reorganization is probably in order.

Alternatively, if an internal function is important enough to need good coverage, it should be pulled out into an internal "library" that exposes the interface explicitly (even if this is just a separate file or folder with limited visibility to the rest of the codebase).

Testing internals is almost always a code organization smell IMO.

Anything as good as this for drum tab transcription?

Which festival? And where?

DRY should really be "DRA" - don't repeat abstractions (which admittedly isn't as catchy).

The problem with DRY is when abstractions with currently identical implementations are given a single interface, even though they're logically distinct.

Then, when those distinct abstractions' implementations need to diverge, you've got a rats nest of references to manually pick through and separate out.

Or worse yet , the mistakenly-shared interface becomes parameterized, leading to a horrible mixing of requirements and concepts that may never be untangled if the original intent is lost to time.

Obviously DRY is great when you can consolidate multiple implementations that are actually a single abstraction, but it can really go off the rails if the motivation for the mantra isn't understood.

Just because it's bad for the planet doesn't mean it's stopping any time soon.

The tragedy of the commons is real, regardless of how irrational or selfish it may seem.

Any decent editor if you don't close it/hard power off the workstation.

It's pretty much always regulatory stuff. Google is a lawsuit magnet so they are extremely paranoid with this stuff to make sure they're not breaking some obscure local law.

Currently in the process of figuring out my next way of making income that brings me real joy thanks to substantial savings from living frugally on FAANG comp.

Lately I've been getting very much into FPV drone videography and 3D printing/design, which are both incredibly gratifying in the way programming used to be for me, but somehow even more so because the process involves mastery of physical matter on top of navigating a high-dimensional abstraction space.

That being said: I don't plan on having kids and enjoy relatively low cost activities, so I don't need to make nearly as much as I did in software to support myself.

I ultimately decided to leave tech when I realized I didn't care much about delivering value and actually just wanted to write beautiful code. The former was a nice bonus for me, but the latter was profoundly captivating.

I'd unfortunately tried to make that mismatch work for too long, and as a result I completely destroyed all of my programming interest via severe burnout.

If this resonates with whoever reads this: please take your passion seriously and protect it. I don't know if I'll ever be able to enjoy coding again, unfortunately.

Someone created this really neat toy for us all to enjoy for free.

They don't owe anyone anything, and certainly don't deserve to be used as an example of what's wrong with an entire discipline.