Happy to continue paying annual for Plex as I think it's worth it, but I'm definitely also installing Jellyfin alongside and starting to look at what really limits swapping completely. Currently that's mostly things like missing PlayStation apps, need for plugins to handle certain features, more complex accounts/auth story, etc. I'm hopeful that with AI development (on a short leash), the Jellyfin core contributors can increase velocity somewhat and really close the gap.

If Jellyfin had a PS5 app I would switch to it instantly, it’s the only thing keeping me on plex right now. Are there other similar apps for streaming your own media to a playstation?

I believe of Plex/Jellyfin/Emby, only Plex has a app for PS5. From what I gather Sony are rather strict about who they allow to develop PS5 apps, and Plex is a "strategic partner" ($$$).

Google Drive reneged on unlimited storage for Education accounts once they realized that universities also contain researchers who need to store huge amounts of data.

Not only did they cut unlimited, they went to insultingly low limits with not much warning after all their nice promises. Moderately large universities ended up with less space per student than the 15GB they give out to anyone for free. It was a pretty bad rug pull.

Massive fraud from abroad didn't help there either. A favorite backup spot for terabytes of pirated media, complete with guides on which schools had good @edu addresses for it.

Hadn't even considered your obvious point, a good one!

I use the non-Pro version for 1080p streaming and have for years. It’s great, does what I want and gets out of the way. Some years ago they were forced by Google to use the standard AndroidTV UI instead of their own custom one, which means it now shows ads on the home screen (a carousel of “watch this on service X”), which are inoffensive enough I haven’t bothered to circumvent them. You can swap to your own custom UI if you want with some ssh futzing.

Added. Thanks!

Maybe they've been working on it, but got scooped?

I don't think that's the case. The numbers in the paper suggest ~92% of the training data comes from pre-existing AI models, including AlphaFold, and they claim things like:

> We largely adopt the data pipeline implemented in Boltz-11 1https://github.com/jwohlwend/boltz (Wohlwend et al., 2024), which is an open-source replication of AlphaFold3

I believe the story here is largely that they simplified the architecture and scaled it to 3B parameters while maintaining leading results.

Nice, I commented looking for this last year: https://news.ycombinator.com/item?id=40588038

Looks like the "covers" need some better instrument isolation, but this is really huge for the music industry.

Accuracy is a useless statistic: give us precision and recall.

Recall 91.5, F1 93.3

I think you need to define which one is the positive, and which one is the negative?

Is AI generated code the positive?

Useless is perhaps a but harsh. It tells you something.

Only if you know the data distribution.

It is pretty easy to get 99.99% accuracy on a dataset that is 99.99% a single class for example.

It tells me nothing because it doesn’t say if they mean precision or recall

It very much tells you something. Accuracy is a measure of overall correctness. Accuracy is something different than precision and recall.

You mean like this one?

http://emilysquotes.com/wp-content/uploads/2014/09/EmilysQuo...

Timeless, ha ha.

If value was actually created every time before it was distributed to shareholders, it wouldn't be nearly as bleak as when the value is instead rapidly extracted from a long-term reservoir to make the numbers go up.