eutectic

We now have games which rely on shooting millions of rays per frame for rendering.

Presumably this means that power is underpriced (potentially for political reasons). Increase prices and any issues should fix themselves.

Resistive losses are related to current (I^2 R), so you need thicker wires to deliver the same power at a lower voltage. I guess it was decided that it wasn't worth the expense.

log 1/p is just -log p

Negative log is the only decreasing function which satisfies f(x * y) = f(x) + f(y). If you have two independent events it makes sense that the total surprisal (information content) should be the sum of the surprisals of the two events.

Another way to see it is as a continuous generalization of the idea that you need n bits to represent 2^n equally likely alternatives.

Language modelling is the objective (if we ignore RLHF). That doesn't mean that interesting kinds of reasoning can't emerge. You could just as easily dismiss humans as being 'biological replicators' which just 'reproduce their genomes'.

I think for O(N^2) transformer inference you need to cache all the activations.

Bacon and ham are highly processed and salted foods.

I don't see the connection.

Yes, but there will always be some necessary energy input.

I think it's more a problem of entropy; You're taking a high-entropy mixture and trying to extract a pure substance. Think sorting red and blue lego.

So is mine, and they made several successful trips which (among other things) makes me think that carbon fiber could be an appropriate material for a submarine. Metallic hulls would also implode catastrophically if not well designed.

The Boeing 787 is primarily made from carbon fiber, including the wings which experience a mix of tensile and compressive loads thousands of times over their design lives. Of course deep sea submersibles occupy a much smaller market where a comparable degree of rigor may not be justified, but I think it's silly that everyone suddenly seems to be a expert in composite and/or submersible design.

Well, obviously going with a more conventional construction would make it easier to come up with a safe design. But I feel like people are going to take the wrong lesson from this accident.

When people say 'carbon fiber is weaker under compressive loads', or 'cylinders distribute compressive forces unevenly', these kinds of issues seem like engineering challenges not fatal flaws. I feel like the problem with the Titan submersible was the lack of respect for safety, not any specific design choice.

Well, in the case of lead, I don't think it was irrational for individuals to feel disempowered. The solution was regulatory and societal change, but how many people want to devote their lives to being PFAS crusaders?

PFAS are used in the manufacture of teflon pans, but teflon pans themselves are not a major source (as I understand it).

And presumably the atmospheric distortion is much less as well.

My experience in London has been quite positive.

I'm not sure it will be so easy to get the ice out if you let it freeze solid.

Seriously? You can't think of a more efficient use of the time and money?

The Haber process is already operating close to maximum possible efficiency, so there are no big gains to be had.

Why would small molecule drugs like these be so expensive? Limited market?

I think it means the analysis accounts for rounding error.

If you change the requirements for an insurance policy you attract a different risk pool. Just like health insurance companies offering gym membership to find healthier customers.

What about electrolysis?

Any idea why utiliy-scale is so much more expensive than automotive?

"never" is a long time.

Presumably they are derived from self-play.

Bees live less than a year, so the number of dead bees can surely only be used as an indicator of changes on around that time-scale or shorter. Eventually fewer bees means fewer dead bees.