The bubble is about everything AI, specifically neural networks. LLMs make all the noise, but audio, video, and imagery are all going all for the ride.

Half the US voted for a convicted felon who explicitly said, out loud, during their campaign, that they'd end democracy. I don't know what the word "criminal" is intended to mean any more.

The reason is twofold: First is mount low-value attacks on the law by breaking the law over things that aren't important. When judges push back, they know who to attack and remove.

The other reason for closing departments with no planning nor warning is to create chaos, and the chaos will give them the excuse to cancel the 2026 elections. Cuz you know they're totally cancelling the elections, it's right there in the playbook.

The whole budget thing is a distraction.

I'll take a risk on responding genuinely, just in case you are also doing so.

The main problem is that the discussion above you is about the specific cost of a specific program. But you've backed out of this discussion with the abstract "ask questions about where your money is going." Yes sir, we're already doing that, we invite you to join us.

The second problem with your argument is that when we need to cut expenditures, we start with the largest ones ones first. That gives us a chance at large savings for a small effort. $60m is a very small expense by US government standards, so you're point doesn't really apply here.

DOGE itself has fallen into this trap, cutting small programs while ignoring the larger expenses.

There are many more problems with Trump's claims as well as DOGE's actions, but I'm just trying to bring you back to the discussion before we tackle those.

The sensor is pretty sweet, 12 megapixel supporting 4k60 and 2k240 video modes. Weirdly there are no specs on the AI processor. I'd view this more as a way to get the IMX500 sensor if you're interested in that. If you're interested in embedded AI, you might look at the Luxonis options, which work with the Pi over USB.

If we follow the joke, then Llama 95 will be released in 2096.

They didn't. That's the store page. Xbox store probably measures the size of all the data uploaded to it as part of the one purchasable item. i.e. If you buy BO6, the Xbox store has 309 GB of content attached to that purchase. The store doesn't know what you "need", only what the publisher has made available.

Depends on which GPU is doing the toasting.

Yes, the one with the LCD display.

TV shows have the opposite incentive, as they generally want to maximize the number of episodes they can generate from the premise. Some shows do this badly, and very clearly work to avoid moving the story forward.

X-Files did it too, and it even happens in video games, like Destiny 2.

When the smallest version of the model is quantized to 4 bits, it only takes 2GB of memory, so yes this should be able to run on actual consumer GPUs.

The A100 is mentioned if you want to "fine tune" the model, which means feeding in new data and training up a modified model. Training a model always needs more time than inferencing with it, but the inferencing can definitely run on a small GPU. (FWIW, my old nVidia 1050 can run small models very well.)

The dataset (The Stack v2) does indeed claim to be 67.5TB, while v1 was 6.4TB. But then it also reports being "~900B tokens," while v1 was "~200B tokens." I don't know what that means, but it might be the source of the 4x number.

Fast autofocus is old news, the question is can it do focus tracking during videos?

That's what we're talking about, right? Auto Focus? Though it's weird they'd mention that without mentioning camera access.

Apple?

It keeps growing cuz nobody can figure out how to exit.