nynx

This is great. May be buying my first car in the next couple months and being able to splurge a little and get an electric vehicle would be wonderful

> It really is one of the best product names in Apple history: Vision is a description of a product, it is an aspiration for a use case, and it is a critique on the sort of society we are building, behind Apple’s leadership more than anyone else.

I agree that “Vision” is a good product name, but “Vision Pro” just feels intensely pedestrian. Why is it called “Pro?” Is there a non-Pro version?

These are “GPUs” in a pretty stretched sense. They have a lot of custom logic specifically for doing tensor operations.

One of exciting outputs of LLMs will be, I hope, to automatically port large software projects to other languages.

Defeats the point of WASI—WASIX seems not to be capabilities based

This is cool. This is for removing nucleotides from the body, rather than repairing damage done by radiation exposure (which probably can’t be done with drugs).

You don’t have access to it.

You’d think maybe they could make the annotations less ugly

If that “data lake” is going to be less than 100 GB, all these people are idiots.

There has got to be a number of fascinating tricks that they're using to support context lengths that long. Shame it's all closed-source.

Eventually, when local compute gets good enough, running LLMs on this kind of system will be far more useful than most internet resources could be directly.

Direct to cell and full starlink internet are basically unrelated services.

You made a correct assertion initially "power (1.5 gigawatts) and energy (power over time)," but then failed to use it correctly. 1.5 gigawatts is a rate, of 1.5 billion joules per second.

I'm a simple man. I see an article from Fernando, I click.

The author confuses sound intensity levels and A-weighted sound pressure. These are, I believe, different physical quantities. Their argument doesn't make sense without that confusion.

How very droll. Why are engineers considered so lowly?

Collecting into a result will short circuit.

Why is it behaving sparsely? There are only dense operations, right?

There are already a million ways to run LLaMA. This doesn't change the issue at all, which is that the weights aren't commercially licensed.

Mostly, yes. LLMs will turn natural language into really whatever form you want.

An application crashing, except in very specific scenarios which don’t apply here, won’t interfere with the rest of the os.

This kind of reminds me of the (very beautiful) beginning of Diaspora, by Greg Egan, in which a nascent machine/human intelligence emerges from a chaotic, abstract virtual machine.

As far as I can tell, they analytically solved the style of ODE used in biologically-motivated neural networks (usually spiking, but not in this case) and then trained a network built from those to do stuff.

Mechanical counter pressure suits pls!

I think there’s a lot of promise in FLF’s designs (especially for in-space fusion propulsion), but they’re moving too slowly.

Watts is a measure of energy over time.

Once they started taking VC money, this is kind of what I expected.

Apologies to Casey if this was in the article and I missed it, but what about nuclear power?

You can have one disruptive rad event every 10 minutes if you compare and correct state between three or five computers at something 100 Hz.

And a more recent computer that isn't designed for rad-environments will automatically be way more efficnent than a rad-hardened one.