ninglor

> “breakout game weights” always look the same regardless of training and model

The prompt told Astral what the input and output layers mean -- that is, how the current state of the game is turned into a vector for matrix multiplication, and how the result of those multiplications is turned into an action to take in-game.

Astral saw that the model was trained on a game where the player is at the bottom of the grid and moves left and right, reward objects live at the top, and there's some not-player-controlled thing moving around in the middle. Astral knows that this type of game is called "breakout."

A model that plays breakout will have meaningful weights for the player being in the bottom row, but not for the player being anywhere else, because the player can't go anywhere else. Astral saw that there were meaningful weights for the player in the bottom row, and nodes representing the player in any other row had low-magnitude noise, so the model had never been trained on the player being anywhere besides the bottom row.

Similarly, breakout has "reward" objects in the top rows, and "there is a reward object at [COORDINATE]" is part of the input layer. Astral saw that the model was trained on reward objects in the top few rows.

Sort of. Phrasing varies, but at least for Idaho it's very common for people who live in Idaho to go on the internet and tell people how awful Idaho is and how you should absolutely not move there. On a related note, for people who do move to Idaho, one of the things they are advised to do as soon as possible is get new license plates so that they're not constantly advertising the fact that they came from out-of-state. Idaho has received a massive amount of immigration from other states and many people are unhappy about this.

It's just the logical contrapositive. "If you have ambition, you find AI useful" necessarily implies "if you do not find AI useful, you do not have ambition."

This sounds like you are deriving an is from an ought.

Is it not conceivable that we won't find a way to do so, and that consumption and economic activity will ultimately decline?

The dairy farmer could be milking chickens!

Yes. Why would US employers require US employees to have those redundancies?

It seems like the employer would be throwing away leverage in the employer/employee negotiation without giving the employer anything they actually value.

Why would US employers require that?

There are many nice places to visit that don't have, and never will have, EV charging.

EV charging sites are still much less ubiquitous than gas stations.

The ~200 mile round trip I did this Monday would have required a 30 minute detour to reach a charger on the way back. Certainly viable, but that's a lot more logistical overhead than just heading out with a full tank in the morning.

The Thalamos link describes a trend from 2023 to 2024. This would not be reflected in a survey from 2022.

This is not meaningfully static content. Look at the charts in TFA. There is a combinatorial explosion of distinct URLs which the crawlers can and do request.

Most people in the US don't drive a new car, and used cars can be had for far less than $40k. An $80k purchase would be just shy of the median annual household income -- anyone who thinks nothing of that has financial resources far above typical. You are in a bubble.

If I want a JDK, I'll ask my package manager to download and install one, and it will be working a minute later. Why would you want to wait another week for something that already exists?

Cool. The gadgets/blueprints paradigm reminds me of Smalltalk or Lisp environments in a way -- you're shipping a messy thing, a packaged version of your own personal, always-evolving environment.

Natural language wrapper around programming. Sandboxing and controlling external interactions to minimize the cost of mistakes. As I see it, the core idea here is facilitating habitability (in the Richard P Gabriel sense) via abstracting software development and distribution in a way that allows non-programmers to engage.

I'm not sure LLMs are quite there with results/$ to fully support a personal computing revolution here. I suspect useful blueprints will largely be wrappers or reimplementations of software that is still primarily created and distributed outside of this environment. But there's promise here, and it's cool.