Eliezer

These weren't Israeli products. If I were Mossad, I'd compromise anything except an Israeli or Jewish-owned product.

They didn't compromise anything that looked Israeli, and targeted other companies.

Q* is also a term from reinforcement learning.

Bad news - there's a 5000-year-old bristlecone pine tree in California. On your philosophy you should go burn it down, I guess? https://www.bbc.com/news/science-environment-40224991

Haven't yet read through it! I probably should at some point.

Linoleic acid (a particular polyunsaturated fat) is possibly implicated as one of the major villains in skyrocketing obesity rates; eg, correlation between obesity and linoleic acid content in body fat, theories about it being a trigger for an archaic hibernation metabolic shift. (Its use has risen hugely over time, along with obesity, but of course you could say the same thing about global warming and pirates.)

That's a hard ask given the actual training runs are multi-month, and have distinguished pretraining and refinement phases.

What an excellent retrospectively obvious point!

Describing this as "on San Andreas" seems deceptive. It's a particular unpopulated location.

Ow.

Prediction markets seem to think 2028. https://manifold.markets/plsdelete/when-will-humans-set-foot...

No, not sarcasm, and I am Eliezer Yudkowsky. I was around on the old Extropians mailing list starting in 1996, and their leadership did not talk like this. Max More (the founder of Extropianism) was a careful thinker then, and I haven't heard anything different about him more recently than that.

"Extropy" is a term that was previously coined by a group of fairly nice people to describe themselves, and so far as I know is being stolen here without permission.

Ah, but running an experiment like that risks it returning an answer you don't like.

I'm saddened to see the honorable name of Extropy and Extropianism, which carefully never descended to this level or anything like it, be stolen and captured by this nonsense.

This is a slightly strange article to read if you happen to be Eliezer Yudkowsky. Just saying.

Grow per capita nominal GDP by a 6%/year level target. (NGDPLT.)

How incredible it seems today that this story could ever have been written by a mainstream reporter. It tells the story of a technology without any attempt at "human color" or "character portraits" to "interest the public". It is written for an adult readership that seems scarcely conceivable today. A lost artifact of lost social technology that could never be reproduced within the modern West.

This reads like the Board 4 are not allowed to say, or are under NDA, or do not dare say, or their lawyers told them not to say, the actual reason. Because this is obviously not the actual reason.

They're usually in the Bylaws. MIRI's Bylaws, iirc 23 years after I wrote them, contain a provision like that.

Every story? Now there's a false title. _Project Lawful_ for example has a sufficiently complex plot with enough opposed character perspectives that you'd have to really work to hammer it into this pigeonhole. Maybe one particular character arc could be said to have this structure.

More generally, any story that changes between character perspectives and evolves in who holds narrative force will have a hard time being analyzed this way.

man, say what you like about my writing, it sure used to be worse

I see they're continuing the campaign to make people dislike something called "AI safety" by redefining it as corporate prudery.

Iiii do not think you should be able to solve the Schrodinger equation with thousands of dimensions in general on a non-quantum computer, what with that being a quantum-mechanical equation some of whose solutions would reflect quantum-hard problems?

Very visceral oof. I don't remember a time when I knew about AI but not about Eurisko.

It's absolutely a side effect of training rather than a bolt-on rule. As I understand and infer: They applied some forms of censorship as thumbed-down in Kenya for $2/hr, and the model updated on some simple pattern that explained those, and learned to talk like a generally censored person - one that resembled text like that in the training data. It learned to pinpoint the corporate mealy-mouthiness cluster in textspace.

And stop there.

Directly productive jobs are often easier to automate than jobs produced by coordination difficulties. Just as we saw sectoral shift from manufacturing to service as manufacturing became easier to automate, we have been seeing a shift from service to bullshit - or so I would wildly guess.

I miss reading things like this.

Too late now, but this sure does sound like what cryonics is for.

Plain old Bing Chat - which I feel a bit better about using because I'm not paying for it and not contributing to the problem that way - used for its original intended purpose of search. Compared to trying out keywords and reading through the results myself, it's faster to ask the question, let Bing read the pages, and let Bing summarize the results to me.

I recently got access to Anthropic Claude, which I don't feel as squeamish about using as I'd feel about paying $20/month to OpenAI and helping them destroy the world, so if there's more that can be done that way maybe I'll find it out soon.