Then I realize I actually don’t have any agents doing anything and I’m not waiting for anything outside of me

They must do this in the minimum number of ASCII characters such that they can be quite confident the other being will know exactly who they are referencing.

No previous common knowledge except common knowledge of the planet Earth and everything that can be logically deduced from that common knowledge.

Now suppose you pick a given person. How would you evaluate whether a proposed string has this property?

If they had a pre-agreed codebook, you’d need roughly log₂(8×10⁹) ≈ 33 bits ≈ 5 ASCII characters to index any human. If they are logically omniscient and omniscient about Earth, but do not have a pre-agreed codebook, how good can they get at minimax? How close to five chars?

It feels like when I get out of bed in the morning feeling groggy and realize that I’ve overslept and have something unpleasant on my to-do list

These are the songs that, if I had found when I didn’t like bluegrass, would have accelerated my conversion.

I still love them all.

Full playlist also on Spotify and YouTube

Note: this is intended for people who already like rock and/or folk music and are trying to see if they like bluegrass. If you mostly listen to pop and/or country, this probably isn’t for you.

• Artemy Kolchinsky
• Aram Ebtekar
• Daniel Chiang
• David Lorell
• John Wentworth
• Fernando Rosas
• Jeremy Gillen

interpret that how you will

I deleted them because I hadn’t phrased them in a way I thought people who disagree would interpret correctly and I was afraid of people disliking me because of opinions I don’t have.

But I think it’s important to be open about my negative feelings towards Anthropic so here I am again on the record saying:

Anthropic is a shitty company because they are trying to build superintelligence and that is putting everyone at great risk;

And yet they are trying to get away with it by pretending to do it safely or more safely than the next guy which is false and absurd.

I do believe many of them are sincerely well intentioned. Alas that is not what counts.

When I tell people I work on AI safety it is not because I think I am going to make the AI safe

People need to stop building the unsafe AI and I cannot make them stop

I do research I think is fun and interesting and it just so happens that maybe my research will be useful to people who want to do AI safely

He responded by talking about the “existential risk of job loss”

Goddammit

“AI just predicts the next token” sounds deflating — until you consider what predicting the next token involves and start to ask if there’s really such a difference between predicting and learning.

The common words in those articles, “just,” “simply,” “only,” are there because the argument doesn’t stand up without them.

https://www.transformernews.ai/p/the-left-is-missing-out-on-ai-sanders-doctorow-bender-bores

I don’t think capitalism is inherently good or bad, it’s just a pretty simple attractor state for societies which has good and bad parts. But I’ve been thinking about how automation changes things.

Democracy is fundamentally about distributed power. When human labor is essential everywhere, workers have power (they can strike, etc).

With full automation, workers have nothing to offer that capitalists need. People could theoretically go live off the land and get by, but they’d be powerless. The automated economy would dwarf anything they could produce, and the capitalists could walk all over them whenever they wanted.

So it seems like you can pick two of {automation, capitalism, democracy}, but not all three. If we want to keep democracy as we automate, investor ownership of production probably has to go.

Questions like ‘what if everyone is an investor’ etc follow and are interesting but I haven’t gotten to them quite yet. Also, this models automated human labor replacement as controllable by owners which is unlikely to be the case in practice.

This is not normally how I feel