Probably LRO vs. Danuri

We don't know the exact answer yet, but IMO, you're two-halves right.

The key difference is, humans aren't "trained" primarily via a clean library of tagged data (which in itself is a learned skill). We experience the world as a more-or-less continual stream of noisy data from our several senses. We self-train how to interpret (and primarily, filter) sensory data; and more importantly, how to derive new training data from it.

To be clear, I'm not saying we have so many neurons to support input filtering (though that's admittedly at least partially true). I'm saying we virtually can't run out of new training data. Every second we're alive, it just keeps coming in.

Except instead of pursuing policies that would help open models at the expense of Big AI, anti-AI people pursue just the opposite - policies that can cripple open source AI but have no realistic ability to meaningfully hurt big AI. It's so short sighted.

Actually, no, what they usually do is much more insidious: fingerprinting rather than watermarking. The fingerprint isn't actually included in the audio, it's included in a database. If they want to tell if the track was generated, they just try to match the fingerprint in the database. Can't filter it out of the track like you can with a watermark.

The two mainstream options (both about half a year old now, we're due for something better) are Ace Step-1.5 and Stable Audio 3 Medium.

Ace Step has full vocals with its music. See here for examples. The downside in my opinion (judge for yourself) is that it has Suno's flaw of sounding too "clean", "mainstream" and "uncreative", except even moreso (Udio was always much better than Suno at this, albeit "less well behaved" - but Udio is out of the game now).

Stable Audio 3 doesn't do vocals, though it has other useful features, like inpainting (great for fixing glitches in recorded tracks for example). In my view, it sounds a lot better. It can also be used to create sound effects. License is a bit more restrictive, though, if you actually care about that, but generally won't affect the average user.

Both are trained on fully licensed data and both are open-weights. So even if the devs released a version that included some sort of watermarking in the future (or someone developed a fingerprinting algo for the existing versions), you could just finetune them to break that.

TL/DR, if you want open source and want to just churn out a full track: choose Ace Step 1.5.
If you want open source and want to do instrumental tracks or to supplement manual work (such as your own vocals) or do audio editing: chose Stable Audio 3.

Current top models are already undertrained by multiple orders of magnitude vs their number of parameters. Not for lack of will or money, but for lack of data - We're already training top models with effectively the entire digitally-available corpus of human knowledge and need about a thousand times more.

That said, the huge Chinese models we've seen lately aren't single models, they're MOEs - Effectively a hive-mind of many smaller models working in concert. Bytedance hasn't mentioned that detail yet but it would somewhat mitigate the training issue. This is still absurdly over-spec'd though. There's simply no way to train it any more than all the rest starving for data.

Not just this month - They also had to revise May and June down by 103k.

It's almost like firing all the people responsible for compiling exactly this data and replacing them with political shills (if at all) was a bad idea. Whodathunkit?

I apologize if "purpose" was a poor word choice. The applicable test is whether or not it's being used verbatim or in a transformative way.

Political campaigns are inherently informative (in a "free speech" sense) and non-profit. While commercial use matters in general, it'd be a seriously uphill battle to claim a profit motive - even if they're spewing disinformation and raking in the cash.

No joke, there are fewer ads on broadcast TV than on (free) YouTube now. Broadcast averages 14 minutes per hour. YouTube is harder to find hard numbers but most sources imply it's around 20 minutes per hour - A whopping third of the total viewing time.

Of course, that depends on having something worth watching on (you'd likely be surprised how many channels you get, but they still mostly suck); and for those of us not using a GooTube-sponsored browser that breaks ad blocking, with uBlock Origin that figure drops to zero minutes per hour.

More practically - Broadcast TV is great for having something random on in the background, and it never pauses to ask if you're still watching. If you're trying to see something specific, though, it's no contest - Streaming is vastly superior in nearly every way.

The lawyers (and IANAL) already understand that, which is why they're stressing the "potential damage to these IPs" angle.

Fair use as it applies to a political campaign has four main tests:
The purpose of the infringing use (must be transformative, which this is);
The nature of the work (creative, which this is, receives less protection than factual);
The amount of material and its importance in the original context (each infringement here is tiny); and
The impact on the market for the original work.

IMO, that last point would be the key issue in this situation. It honestly passes the first three tests with flying colors, but it also very likely adversely affects the market for the original content. A third of the country actively avoids anything and everything related to the current administration. If people are plausibly likely to believe Nintendo / Shueisha / Publisher-X don't object to these uses of their material - And the general public have no clue this battle is even going on - They will stop buying it.

We can supplement human vision with those technologies. Currently less than 1% of US cars have any LIDAR capabilities whatsoever, and you'll find a substantial number don't even have backup cameras (only required since 2018).

As for how useful they are, I respectfully disagree. The first thing I, and many others, do when driving a car with a million and one distracting "safety" features is disable most of them. Yes, I know perfectly well I'm drifting out of my lane, now stop nagging me so I can focus on avoiding the real danger a quarter mile ahead!

At face value, very much so. I'm not going to panic until the next decade of appeals plays out, though.

This first round was against defendants who didn't even bother showing up to court (and let's be honest, there really wasn't any viable defense they could have offered here). Thanks to this overly broad injunction, the deep pockets of every ISP and cloud hosting provider in the US now has standing to get involved - And they will, since the precedent this sets could cost them real money in the future.

"Humans of course can drive with just eyes, so there's that proof of existence"

If Waymo's argument is valid, then humans shouldn't be allowed to drive.

The problem is therefore either solvable purely via optical methods, or this is merely a prelude to banning all of us from driving. And if that's the real end goal, why are we messing around with self-driving single- (or low-) occupancy vehicles rather than investing in viable mass transit?

Not the GP, but I suspect an awfully lot of us have zero interest in "owning" the second most expensive possession most of us have, if we're not allowed to operate them ourselves.

There's 8 billion people on the planet. Are you saying 7 billion a year do get drunk?

Out of those 8 billion, 7,999,999,999 did not make as stupid a comment as you did. Let's celebrate!

That may or may not be a defect. Human eyes are also not seeing in all directions at once, unlike Tesla's cameras.

And it has nothing to do with the CEO's lie that cameras cannot be safer than humans, when it is the automation that makes automated cars safer. Mount a lidar to your roof. Are you now a safer driver? No. It is the automation which is safer, not the lidar alone.