topranks

He was knocking crypto from the very start

Such an odd world where someone like Patrick Boyle is considered “left wing”.

Hahaha haven’t heard that one in a long time. Cracked me up.

Crypto is in no way better money.

It’s slow and expensive and doesn’t scale.

The real answer is an alternative that relies on a centralised provider - or set of them - who use traditional databases and the like, but do things in a modern efficient way and don’t charge the same fees as the current card providers do.

Yeah I’d love another mini, but alas.

I remember Ubuntu were going for that “dock it and it’s a full computer” idea about 10 years back with their failed phone OS project.

You might be anti-ai in the sense you aren’t comfortable with all your data being shipped back and forth to a third party.

Yeah I don’t think they are over-priced, last I checked all those companies are still losing money hand over fist.

I guess the question is more he doesn’t want people catching up by doing cheaper training (through distillation or otherwise), and he definitely doesn’t want companies to spend their money training these models and then _giving them away_, which fundamentally undercuts their commercial model and any way to claw back their investment

Also all the bad food choices me and my fellow stoners make when we’re high.

Yeah but messaging apps are really only useful if there are lots of people on them to message.

So in the real world a relatively small number of providers, WhatsApp, Signal etc, are in a position where all your friends are going to be on them. And those are the ones likely to be named and told they need to implement image scanning/review.

Really hope they aren’t forced to do that.

Apple TV one of the few streamers with decent quality audio.

We don’t need a “loudness war” for TV and movie content.

You’re way off here.

Conflating DCT-based compression of audio data (like MP3) to dynamic-range compression of an audio signal (as done on an audio compressor during production) shows how little your grasping the problem.

It doesn’t work very well for TV in my opinion. Ads still sound louder as they are mastered differently.

I’ve no skin in this game, and no desire to ever see or hear ads. I just hope we don’t kick off a “loudness war” for TV/Movie audio by mandating the average volume across entire programs has to hit some high level like modern music or ads have.

Uh huh.

What people are now talking about here is to normalise the “average volume” (by using compression to increase it, or just lowering the volume to decrease).

The problem is the “average volume” of 30 minutes of video, depending on the content, could be quite low.

So matching the ad to that level might not work. The nightmare response we might get is the average volume of TV/movies will be pushed to the limit in future, removing any dynamic range or impact. To comply with the rules for ads. Sigh.

You “turn it up” by using dynamic range compression. Which is fundamentally altering the work.

It’s right and proper YouTube do not do this to people’s work.

Because content creators have a right to produce audio how they want.

If it’s a video of someone speaking in a quiet voice it shouldn’t have to have an average volume level of a new Metallica video. Forcing people to use heavy dynamic range compression to get it there would not be good.

Replaygain works reasonably well for modern music, as do the LUFS units.

The problem with video content is there can be extended very quiet parts. If you try to bring the “average” volume of an ad to the same “average” volume as the last 30 minutes of a drama it could end up being insanely low. The average level of the drama being low due to long quiet parts.

Not always the case but this problem isn’t as simple as it might seem.

My main hope is that this doesn’t kick off a “loudness war” for tv/movie content on streaming in an effort to get its average volume up as high as the ads.

But the training expense is part of their costs!

The question is can they just stop training at some point, fixing the models in time, and still have a useful product.

It definitely costs them a lot.

Implementing v6 without doubt saves money for ISPs. Especially in the CGNAT game as those boxes aren’t cheap.

Exactly. And because the infra was never public in the US, and thus no “privatisation with rules”, it’s been a bit of a disaster.

Indeed.

This is a funny legacy of how telecoms was run and regulated in different countries.

Now the US has the least consumer choice, and shitty ISP practices from exploiting peering relations to selling user data.

I often said net neutrality was fine, but it was an answer to a problem created by the legally-enforced monopolies in the US.

Was that real DIA??

I’ve seen this commonly from them with residential service. They mostly give you a private v4 address behind CGNAT in addition to IPv6. If you want a public v4 they put you on their old network which has no v6.

A lot of web scraping. The ones that got KimWolf’d were/are doing a lot of DDoS (SYN floods etc).

It’s very tricky because the IPs are all on normal user ranges you can’t block without blocking those users.

The company behind this blog - spur.us - offer some paid services I think. There is also this project from Wikimedia which uses that data to produce more manageable lists:

https://gitlab.wikimedia.org/repos/sre/CIDERGRINDER

VPN ranges at least are obvious so that’s different.

Tor less so but it doesn’t seem to be commonly used for this kind of abuse.

Absolutely.

Bear in mind the scrapers wouldn’t need to use these proxies were they not being blocked by the sites they are scraping. So it’s being used to evade blocks.

For some content the level of scraping is outweighing real users, driving up costs and pushing them towards more closed models.

Wikipedia for example make content available free, if you start hammering the site they will rate limit you to keep the lights on. If you need the data fast in bulk they have a paid program to get it without scraping. But some prefer to neither adhere to reasonable request limits nor pay for their use of the infra; instead they choose to pay these grifters to avoid the rate limits.

Nah it’s messing up peoples home internet, and massively abused to perform denial of service attacks and scraping of web content by large AI companies who are otherwise blocked.

Your vision of a randomly-routed mesh internet overlay is also not very scalable bandwidth or latency wise.

I connect every few months (with cable) to check for firmware updates and the like. Otherwise agree it stays offline.

This is to frustrate those using distillation techniques to train their own models right?

There is, see my other reply.

CDs are able to store much louder tracks than can be cut on a record. The technical reason things on CD got louder is because they could.

CD has better dynamic range, sure. But a CD is also able to represent a signal with _much less dynamic range_ than it’s possible to cut on a record.

It was the move to digital that facilitated the loudness war.

In modern years it’s been fairly common for masters to vinyl to be less compressed than the CD release, for the simple reason vinyl has more limitations.