Autism intensifies

I’m extrapolating from history.

15 years ago people made fun of AI models because they could mistake some detail in a bush for a dog. Over time the models became more resistant against those kinds of errors. The change was more data and better models.

It’s the same type of error as hallucination. The model is overly confident about a thing it’s wrong about. I don’t see why these types of errors would be any different.

multiple clones

Why would you do this to yourself?

The benefit is that you have everything collected in one place. You can jump between any of your local branches, and there’s no confusion about which state the branches are in.

If you have multiple clones, then there’s the risk that you’ve forgotten to sync main in all your different clones.

Then there’s also the problem that all the generated binaries will be out of sync. You still have 5 copies of each binary.

Most improvements in machine learning has been made by increasing the data (and by using models that can generalize larger data better).

Perfect data isn’t needed as the errors will “even out”. Although now there’s the problem that most new content on the Internet is low quality AI garbage.

Ideally, youtube won't be natively encoding the ads into the videos, because that would be a nightmare

I’m afraid this is what they’re going for.

It’s only a matter of time until YouTube stops that as well.

Database of ad timestamps like sponsorblock only works if ads happens at the same timestamps (and are of equal length). This is not necessarily the case.

The only reliable way I can come up with is a database of ads to look for, but that can be huge to accommodate for all possible ads. There’s also risk of false positives (risk of skipping video when there are no ad).

I’m not sure if a sponsorblock like solution will work. Sponsorblock is entirely reliant on timestamps provided by users.

A similar solution for YouTube’s ads will only work if the ads always happen at the same timestamps and have the same length. This is not necessarily the case, as ads can happen at any point.

They don’t need to do any extra transcoding. It’s not that costly to stitch videos together. If done at specific strategic locations, it’s like copying a text file into another.

It’s maybe doable, but figuring out where ads start and end in a video file isn’t necessarily trivial. The app must know what to look for, and YouTube might try to obfuscate it to make it harder.

That’s the point. They want to stop people bypassing ads by using alternative front ends. If they succeed with server side ads, then it’s going to be difficult to block ads. Maybe not impossible, but difficult.

If you think you’re the center of the world then all your opinions will feel centrist

I think assets like app icons are ok. They rarely change, and are often quite small. It’s convenient to have those kinds of things bundled together with the code.

Linux has developers. It just needs more desktop users.

He’s also promising a fully self drivable car by the end of 2017

Well that makes him an even greater asshole.

I don’t think percent works like that.

Bitcoin transactions are public. Anyone can view your transaction history if they know your wallet address. It’s not a good option for privacy.

Also, it’s not true that it hasn’t seen downtime. It has happened at least once in its early days due to a bug. Also, there has been many times where it taken more than an hour between blocks. This is more to its probabilistic nature.

Most cryptocurrency transactions are public information.

Monero is an exception.

I know, I read it because I wanted to know too know if it was addressed