7e

Amazon Leo has about 350 in orbit now, on their way to 7,500.

That is actually better in every way! Assuming the time zones are 2D so the dawn is local. It would probably cut down on cancer rates, too, by harmonizing circadian rhythms.

Boo, in the age of AI we need two dimensional time zones which skew the clock a few seconds every night, continuously, so the sunrise is always at 6 am. Humans were made to rise with the sun.

Ownership means nothing without control.

GLM-5.2 seems capable. It’s just much slower than Opus.

[flagged]

These are laws passed by democratically elected legislatures. It’s not a legal issue, it’s a political one. Ostensibly, the will of the people. But these wimpy geeks passively-aggressively protest by arguing about code online, instead of engaging in real political activism.

Nth order effects are unpredictable. You can only observe them after the fact. Systems thinking is a debugging methodology, not at all a crystal ball. Witness: your comment makes a bunch of predictions that are very unlikely to be true.

Nth order effects are unpredictable. You can only observe them after the fact. Systems thinking is a debugging methodology, not at all a crystal ball.

You also can’t build a house which violates building or electrical codes, or drive without a license. These are safety and security protocols and the digital realm now has them.

Mobile devices are secure and that’s why they’re not infected with malware, like any Windows machine. This is why Android is the host of 98% of all mobile malware and iPhone is not.

You have the freedom to make your own insecure devices which don’t have any trusted or secure elements. Go for it! Take your GNU and go wild.

No, you can reliably attest public source builds of critical software for the ultimate in transparency. That even includes models running on GPUs. Combine that with blind tokens and you get trusted, anonymous identity verification.

Trusted computing fixes this up to the analog hole. Which is as much as you can expect.

Trusted computing solves this problem handily.

Pretty obvious stuff; see Terminator for the conclusion (SkyNet). Or the Matrix. We really need more work on model alignment, trustworthiness, and control.

Everyone is doing this already. All the frontier model makers, as well as a number of startups which are trying to be the IDE for the center of the octopus. Nothing about this is novel.

Companies like Anthropic and OpenAI need to sponsor open source projects by giving them free agent credits. Otherwise, bad actors can just outspend and totally overwhelm the somewhat dim and very overworked set of human maintainers. Humans in software are obsolete, full stop.

My bet is that Anthropic will be exposed as openly evil within the next five years--even if they aren't even secretly evil now. That's the arc of the sociopathic corporate brain, every time.

This is endemic with open source. Nobody owns the hard bugs, nobody does the systems spelunking, and there is little power to make cross-cutting changes. This is why I use a Mac. Not perfect, but better than that no-accountability midden-heap that is Linux.

Maybe AI coding agents will make the situation better, but because open source maintainers are too dim to understand the complex changes the AI makes, and too poor to have their own AIs to help them, they won't take the changes. I make improvements to open source but am forced to keep them to myself.

Tesla is a big part of this. They ship their headlights misaligned from the factory, so they point right into the drivers eyes. Tesla has no quality culture at all. A bunch of wankers.

Any market that Anthropic suddenly thinks is valuable will silently and suddenly be off limits to you. They will train their model on your prompts, and then become your competitor.

Let the price gouging begin.

A "job opening" is not a job. It's an aspirational advertisement.

Further, the graph shown is pretty noisy and I'm not sure the upward move which counters the downward trent is statistically significant.

The market should consider this a huge negative: SpaceX is renting out their compute because they have failed to make use of it themselves. This calls into question whether they have any talent in xAI at all.

That's a budget thing. Claude is suffering from huge demand and they're pulling out all the stops to try to keep the lights on: terse tokens, lobotomizing Claude six ways from Sunday, aggressive batching, the works.

Just pretend that it is the next step update when training. You didn’t train your model to step=inf, I hope?

More evidence that the original Transformer authors didn't really know what they were doing, but they did have access to more cheap compute than anyone else.

They don’t consider risk of ruin and that is where this calculus falls apart. The reward does not reduce the risk of ruin, which increases with blast radius. YOLO!

Curious how this compares to the keyboard.io and of course, the incredible Svalboard.

systemd is great all around. Don’t listen to the boomers complaining about it because their cheese was loved.

If Judaism were smart, they would label AI as their Messiah finally come. Then at least some prediction would finally come true in a major religion.