LessWrong 2.0 Reader

I reflexively downvoted this, so I feel obliged to explained why. Mostly because it reads to me like content-free word salad repeating the buzzwords like Solomonoff induction, Kolmogorov complexity and Occam's razor. And it claims to disprove something it doesn't even clearly define. Not trying to impose my opinion on others here, just figured I'd write it out, since being silently downvoted sucks, at least for me.

Fixed the code blocks for you. The trick was to press CTRL+Shift+V to make sure that your browser doesn't try to do any fancy formatting to your code. Sorry for the inconvenience.

What this means is for the Jacobian is that the determinant tells us how much space is being squished or expanded in the neighborhood around a point. If the output space is being expanded a lot at some input point, then this means that the neural network is a bit unstable at that region, since minor alterations in the input could cause huge distortions in the output. By contrast, if the determinant is small, then some small change to the input will hardly make a difference to the output.

This isn't quite true; the determinant being small is consistent with small changes in input making arbitrarily large changes in output, just so long as small changes in input in a different direction make sufficiently small changes in output.

The frobenius norm is nothing complicated, and is really just a way of describing that we square all of the elements in the matrix, take the sum, and then take the square root of this sum.

An alternative definition of the frobenius norm better highlights its connection to the motivation of regularizing the Jacobian frobenius in terms of limiting the extent to which small changes in input can cause large changes in output: The frobenius norm of a matrix J is the root-mean-square of |J(x)| over all unit vectors x.

I agree with this completely. I would add that the "Lists" feature changed the way that I use Twitter for the better. I now have lists for each of my core interests and & topics -- Economics, Aviation, San Francisco, etc. -- and turn to those when I'm in the mood for specific information relevant to those areas. It helps that all of the tweets are in the context of a specific topic, versus the normal feed, where everything is reverse chronological and jumps from topic to topic.

I should note, I don't know how to argue persuasively for faith in solomonoff induction (especially as a model of the shape of the multiverse). It's sort of at the root of our epistemology. We believe it because we have to ground truth on something, and it seems to work better than anything else.

I can only hope someone will be able to take this argument and formalise it more thoroughly in the same way that hofstadter's superrationality has been lifted up into FDT and stuff (does MIRI's family of decision theories have a name? Is it "LDTs"? I've been wanting to call them "reflective decision theories" (because they reflect each other, and they reflect upon themselves) but that seemed to be already in use. (Though, maybe we shouldn't let that stop us!))

The target audience for Hazardous Guide is friends of yours, correct? (vaguely recall that)

A thing that normally works for writing is that after each chunk, I get to publish a thing and get comments. One thing about Hazardous Guide is that it mostly isn't new material for LW veterans, so I could see it getting less feedback than average. Might be able to address by actually showing parts to friends if you haven't

So I'm making lists of things I might like (cooking, reading, improv, etc) and I'll try those

This comment is a bit interesting in terms of it's relation to this old comment of yours (about puzzlement over cooking being a source of slack)

I realize this comment isn't about cooking-as-slack per se, but curious to hear more about your shift in experience there (since before it didn't seem like cooking as a thing you did much at all)

Promoted to curated: I think this post is describing a real error mode that I think many people benefit from identifying. I also particularly appreciate that this post tries to provide concrete evidence in the form of personal experience (though obviously externally verifiable evidence is even better, but also even harder to come by). I think a frequent error mode for posts on LW is to not be grounded sufficiently in even personal experience.

META: The code block editor wasn't very friendly and ate up all of my tabs. I'm working on better formatting, and this'll probably end up being a post on my own blog later on, which will hopefully also have things like syntax highlighting.

Huh, 95% is quite extreme. But I realise this probably also solves the problem whereby if the people I'm interested in comment on *someone else's* wall, I still get to see it. I'll try this out next week, thx.

(I don't get to be confident I've seen 100% of all the interesting people's good content though, the news feed is fickle and not exhaustive.)