Eliezer

7 ms·

31.
by Eliezer 2y ago
They didn't compromise anything that looked Israeli, and targeted other companies.
32.
by Eliezer 2y ago
Q* is also a term from reinforcement learning.
33.
by Eliezer 2y ago
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
34.
by Eliezer 2y ago
Haven't yet read through it! I probably should at some point.
35.
by Eliezer 2y ago
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
36.
by Eliezer 2y ago
That's a hard ask given the actual training runs are multi-month, and have distinguished pretraining and refinement phases.
37.
by Eliezer 2y ago
What an excellent retrospectively obvious point!
38.
by Eliezer 2y ago
Describing this as "on San Andreas" seems deceptive. It's a particular unpopulated location.
39.
by Eliezer 3y ago
Ow.
40.
by Eliezer 3y ago
Prediction markets seem to think 2028. https://manifold.markets/plsdelete/when-will-humans-set-foot...
41.
by Eliezer 3y ago
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&#x2
42.
by Eliezer 3y ago
Ah, but running an experiment like that risks it returning an answer you don't like.
43.
by Eliezer 3y ago
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.
44.
by Eliezer 3y ago
This is a slightly strange article to read if you happen to be Eliezer Yudkowsky. Just saying.
45.
by Eliezer 3y ago
Grow per capita nominal GDP by a 6%/year level target. (NGDPLT.)
46.
by Eliezer 3y ago
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
47.
by Eliezer 3y ago
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.
48.
by Eliezer 3y ago
They're usually in the Bylaws. MIRI's Bylaws, iirc 23 years after I wrote them, contain a provision like that.
49.
by Eliezer 3y ago
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 part
50.
by Eliezer 3y ago
man, say what you like about my writing, it sure used to be worse
51.
by Eliezer 3y ago
I see they're continuing the campaign to make people dislike something called "AI safety" by redefining it as corporate prudery.
52.
by Eliezer 3y ago
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
53.
by Eliezer 3y ago
Very visceral oof. I don't remember a time when I knew about AI but not about Eurisko.
54.
by Eliezer 3y ago
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 explain
55.
by Eliezer 3y ago
And stop there.
56.
by Eliezer 3y ago
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
57.
by Eliezer 3y ago
I miss reading things like this.
58.
by Eliezer 4y ago
Too late now, but this sure does sound like what cryonics is for.
59.
by Eliezer 4y ago
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 thr
60.
by Eliezer 4y ago
Literally retraining on inputs and outputs without doing anything clever, does not to me seem to deserve quite such a fancy name. How do we end up with a smaller model if the dataset formats are exactly similar, except in some other special

More ›