philip_b

Karma: 532

philip_b 19 Sep 2021 13:39 UTC
3 points
on: Kids Moving Pictures
You have very cute kids :)

philip_b 10 Sep 2021 11:46 UTC
6 points
on: The Best Software For Every Need
Software: The free version of PDF Viewer Pro by PSPDFKit (https://apps.apple.com/de/app/pdf-viewer-pro-by-pspdfkit/id1120099014#?platform=ipad) for iPad

Need: A program for iPad for reading PDFs and writing/making notes in them using apple pencil. I use it to attentively read math textbooks and solve exercises in them, read psychotherapy self-help books and fill in worksheets in them, read difficult academic articles and write notes for myself.

Non-needs (i.e., if you need these, maybe you should disregard my recommendation, because I don’t use these): signing documents and sending them to people, rapidly giving feedback to people in the form of annotating their pdfs, collaborative anything, reading fiction without annotating or taking notes or anything, using a dictionary to look up unknown words.

Other programs I’ve tried: Xodo, Liquid Text, MarginNote3, Flexcil, Flexcil 2.

Pros:
- I can easily add blank pages or pictures, remove pages.
- Very good tools for writing or drawing inside a document.
- It quietly saves my notes in the pdf file. I use MobiusSync, so the changes get synced to my computer, and there are no problems with seeing the notes in other PDF viewers.
Cons:
- Doesn’t open djvu files. (but neither do any other of the compared programs)
- A sketchy privacy policy written in legalese, so I can’t even understand what data it uploads and whether it uploads my documents.
Xodo provides all the same functional, but there are some bugs that make it worse than PDF Viewer Pro by PSPDFKit:
- Sometimes I try to add a blank page after, let’s say, page 200, but instead it gets added after page 201 or page 199.
- Erasing handwritten scribbles is buggy.
- Tools for writing or drawing inside a document are much less pleasant to use.
- Sometimes drawing becomes shifted relative to where I am touching the screen with apple pencil.

philip_b 20 Aug 2021 13:41 UTC
3 points
on: Covid 8/19: Cracking the Booster
I play MTG but I don’t understand the Oko, Thief of Crowns meme. Can you explain?

philip_b 8 Aug 2021 11:46 UTC
4 points
on: Rage Against The MOOChine
There are a lot of very good resources to learn ML that are accessible for free. Here I list some of them.
- Introduction to Applied Linear Algebra: Vectors, Matrices, and Least Squares—Boyd 2018. It’s a not-very-deep (but much deeper than your average MOOC) linear algebra textbook with the focus on ML and ML-adjacent applications. For instance, it covers k-means clustering, least squares data fitting, least squares classification. Boyd is known for being the authour of the best introductory textbook on convex optimization, hence this textbook is probably good as well.
- Linear algebra and learning from data—Strang 2019. This one assumes some knowledge of linear algebra and teaches how to use all that on contemporary computers efficiently and also it teaches many ML and ML-adjacent methods using that knowledge, including (stochastic) gradient descent, neural networks, backpropagation, some statistics. Strang is known for being the author of one of the best linear algebra textbooks “Introduction to linear algebra”, hence this textbook is probably good as well. There is also a free MIT opencourseware course https://ocw.mit.edu/courses/mathematics/18-065-matrix-methods-in-data-analysis-signal-processing-and-machine-learning-spring-2018/ that follows this textbook.
- The elements of statistical learning. Data mining, inference, and prediction 2nd edition—Hastie and Tibshirani 2008. This one covers A LOT OF ML, although all of it is pre-neural-network-revolution. Most exercises are of form “prove this theorem” rather than “implement this in code”.
- Pattern recognition and machine learning—Bishop 2006
- Probabilistic Machine Learning: An Introduction—Murphy 2021
- Deep learning—Goodfellow, Bengion, and Courville 2016
- AI: a modern approach 4th edition—Russell, Norvig 2020
Some of these are free but most are not. However, you can pirate them off libgen.

philip_b 21 Jul 2021 11:29 UTC
18 points
on: My Marriage Vows
Security mindset, here we go.

I will never set out to deceive my [spouse] on purpose without [pronoun]’s unambiguous consent, without exception.

Perhaps you two will realize that it’s better for you if your partner deceives you in some class of situations for stupid-monkey-brain reasons. I guess you can patch the agreement then by consenting to such deception.

Another potential failure mode: you will think that you’re not deceiving your partner in some area while actually deceiving them. According to “The Elephant in the Brain”, this is probable. From having read the vows, I am unsure how they are supposed to treat such a situation. Also, maybe in such a situation your partner will start despising the vows thinking that it’s unfair that you’re better at deceiving-without-knowing-that-you’re-deceiving than them and that because of these, the vows are better for you. Or it could be vice versa.

Also, for elephant-in-the-brain reasons, I think it would be good to make the agreement such that there would be some incentives to not act against your partner via unconscious elephant-in-the-brain mechanisms.

The only exception to the latter is when this information was given to me in confidence by a third party as part of an agreement which was made in compliance with all Vows

This might make third parties less likely to tell you important things, since you might tell your spouse.

I will also never withhold information that [pronoun] would in hindsight prefer to know.

What happens if someone learns about this agreement of yours and thus kidnaps and tortures your spouse until they tell the kidnapper all your passwords or something like that? Suppose your spouse didn’t know your passwords before this event and now they’re on the phone and asking for your passwords. Are you obligated to give them your passwords because right now they prefer to know? Or are you free form the obligation because before this actually happened, both you and your spouse would prefer that you don’t give passwords?

Another thing, I think this whole agreement models you as a rational agent with a utility function. And it might become bad for you if you start thinking about yourself another way, e.g. if you start seeing yourself less agentic, less rational, with more monkey-brain bugs.

philip_b 15 Jul 2021 12:53 UTC
5 points
on: Experimentation with AI-generated images (VQGAN+CLIP) | Solarpunk airships fleeing a dragon
That is super awesome and surprisingly good.

philip_b 7 Jul 2021 7:19 UTC
9 points
on: Chess and cheap ways to check day to day variance in cognition
http://www.quantified-mind.com/ contains many cognition tests and games which can be done quickly.

philip_b 11 Jun 2021 11:38 UTC
3 points
on: Would you like me to debug your math?
Does your offer stand for learning math instead programming something? Some possible examples are:
1. Solving exercises in the end of a chapter
2. Given a chapter from a textbook, redoing it on my own for a more general case or for a slightly different case .
3. Reading some proofs and understanding them.
4. The whole loop of reading a chapter, understanding it, looking up things I don’t understand, solving exercises, proving things I want to prove on my own, adding notes to my notes database.

philip_b 23 May 2021 12:26 UTC
1 point
on: crabman’s Shortform
Please recommend me cheap (in brain-time and text length) ways to represent different levels of confidence in English texts and speech. More concretely, I want English words and ways to use them in sentences without modifying the sentence’s structure. Examples:
- This is a reason to think that SGD-like algorithms can’t learn it. → This is a reason to think that maybe SGD-like algorithms can’t learn it. // sounds kinda bad, I wouldn’t write it like that for people who are picky about style; also, what other words work here instead of maybe?
- because the inputs and the parameters are noisy → because likely the inputs and the parameters are noisy // seems unclear
Non-examples:
- SGD-like algorithms can’t learn this circuit because the inputs and the parameters are noisy during training. → It’s plausible that SGD-like algorithms can’t learn this circuit, because it’s plausible that the inputs and the parameters are noisy during training. // the structure of the rest of the sentence had to be modified, also I had to spend brain-time to think about how to formulate this. also it’s very long

philip_b 14 May 2021 20:19 UTC
4 points
in reply to: johnswentworth’s comment on: Understanding the Lottery Ticket Hypothesis
The tangent space at a point a is tangent to what manifold?

philip_b 14 May 2021 9:16 UTC
4 points
on: Understanding the Lottery Ticket Hypothesis
Nice article. It was easy to read, the explanations were mostly clear. I’ve finally learned what LTH is, yay. I have a few questions.
1. What is a subnetwork of a convolutional neural network? Is it the function obtained by zeroing some non-bias parameters? Does the same also hold for resnets, i.e. all skip connections remain?
2. Can you add a qualification in the very beginning saying that all this applies to multilayer perceptrons and CNNs in supervised learning rather than all NNs?
3. You often use the word dense, and sometimes I don’t understand if you mean a fully connected layer or an unpruned (fully connected or convolutional) layer. Can you clarify it for all occurrences where it might be unclear?
In some follow-up work, Frankle and Carbin also found it necessary to use “late resetting”, which means resetting the weights of the subnetwork not to their original values from initialization but to their values from 1% − 7% of the way through training the dense network.

Do you mean weight that the network had after 1%-7% of all iterations have happened? Or do you mean moving 1%-7% of the line segment from the initial weights to the weights when the training was stopped?

philip_b 10 May 2021 15:10 UTC
1 point
on: Open and Welcome Thread—May 2021
Is Ziz, the author of https://sinceriously.fyi/, a blog about saving the humanity from unfriendly AI, hating on MIRI and CFAR, and gender, banned on LessWrong? If yes, why and is it for reasons similar to [the ban of ialdabaoth][https://www.lesswrong.com/posts/uwBKaeQzsvkcErmBm/ialdabaoth-is-banned)? If he reasons are similar , then why was there no site-wide announcement like with ialdabaoth?

philip_b 6 May 2021 11:02 UTC
2 points
in reply to: ChristianKl’s comment on: Quinn’s Shortform
http://acritch.com/ai-berkeley/
If you get into one of the following programs at Berkeley:
- a PhD program in computer science, mathematics, logic, or statistics, or
- a postdoc specializing in cognitive science, cybersecurity, economics, evolutionary biology, mechanism design, neuroscience, or moral philosophy,
… then I will personally help you find an advisor who is supportive of you researching AI alignment, and introduce you to other researchers in Berkeley with related interests.
and also

While my time is fairly limited, I care a lot about this field, and you getting into Berkeley is a reasonable filter for taking time away from my own research to help you kickstart yours.

philip_b 6 May 2021 10:13 UTC
1 point
in reply to: ChristianKl’s comment on: Quinn’s Shortform
Critch’s written somewhere that if you can get into UC Berkeley, he’ll automatically allow you to become his student, because getting into UC Berkeley is a good enough filter.

philip_b 3 May 2021 14:18 UTC
12 points
on: The Loudest Alarm Is Probably False
For a data point, after having read the first paragraph I looked for examples and counterexamples. I found 1 example where a person strongly worries they are X while actually they’re anti-X (just like orthonormal describes), 2 examples where a person strongly worries they’re X and they’re indeed X, and 3 examples where they worry they’re X but actually they’re normal, i.e. between X and anti-X.

philip_b 27 Apr 2021 11:37 UTC
1 point
in reply to: Raj Thimmiah’s comment on: Is there software for goal factoring?
I don’t do goal factoring regularly, so I haven’t used yed that much. For very small graphs I find graphviz better. If for some reason I need to draw a large graph and I don’t need it to look very good, I’ll probably go for yed again.

philip_b 19 Apr 2021 15:43 UTC
6 points
on: Parameter count of ML systems through time?
Since you wrote about OpenAI’s “AI and compute”, you should take a look at https://www.lesswrong.com/posts/wfpdejMWog4vEDLDg/ai-and-compute-trend-isn-t-predictive-of-what-is-happeningfor.

philip_b 30 Mar 2021 9:53 UTC
2 points
in reply to: Pato Lubricado’s comment on: Log-odds (or logits)
See Jach’s reply.

philip_b 30 Mar 2021 8:02 UTC
1 point
in reply to: Gurkenglas’s comment on: Gurkenglas’s Shortform
https://activitywatch.net/

philip_b 23 Mar 2021 6:52 UTC
3 points
on: A Retrospective Look at the Guild of Servants: Alpha Phase
You abused passive voice too much. In the following examples, you made it unclear who suggested these ideas and whether the guild council agrees with them and plans to implement them.

For the Beta Phase, it is recommended that the Guild leadership structure be reorganized such that each Guild Council member is assigned a specific administrative role, and that participation and contribution be compensated either monetarily or with ownership equity.

It was recommended that all courses begin with a questionnaire to determine each student’s background knowledge and competency in the course domain.

It was suggested that a professional fashion expert be paid to give feedback at specific junctures to avoid groupthink and also enhance perceived credibility.

It was suggested that this class could benefit from some social manipulation games or inter-cohort competitions, potentially with prizes.