Trace of a Linear Operator We define the trace as the sum of diagonal elements of a matrix: Lemma : If and are square, then . Proof…
Modified: March 16, 2022.
There are three main approaches to moral philosophy: [ utilitarian ]ism: you should feed a starving person because it will increase 'global…
Modified: June 07, 2021.
These days we think a lot about using data to train large [ language model ]s. But there's only so much data in the world; eventually we'll…
Modified: October 27, 2022.
I didn't have a good intuitive understanding of the social landscape of being a researcher (and joining a [ research community ]). When…
Modified: February 25, 2022.
If you and I agree of our own volition to exchange X for Y, this implies that we both believe we are gaining value in the trade. If one of…
Modified: February 22, 2022.
The core of the transformer architecture is multi-headed [ attention ]. The transformer block consists of a multi-headed attention layer…
Modified: February 13, 2023.
What does the computational profile of a transformer vs a similar RNN look like? First, the transformer. Let's take the LLama 6.7B model…
Modified: October 04, 2023.
In developing intuition about [ transformer ]s it's useful to think about specific primitive operations that can be implemented by a small…
Modified: February 13, 2023.
Incorporating explicit memory and retrieval seems pretty clearly like the next frontier in language modeling and AI more broadly. We have…
Modified: September 03, 2022.
According to this reddit post , one of the main takeaways of functional analysis is that the right way to interpret the 'transpose' of a…
Modified: November 06, 2020.
SSC link: How general is this phenomenon? You have a belief Your belief colors your perception of something that doesn't inherently…
Modified: March 12, 2021.
Sasha Chapin describes trauma as a 'splitting off' of difficult or painful experiences as memories that the mind tries to avoid accessing…
Modified: October 27, 2024.
Modified: .
A pitfall with relying too heavily on rational deduction is that lots of logically 'true' conclusions are unimportant, or worse yet…
Modified: February 15, 2022.
(notes loosely based on the Berkeley deep RL course lecture ) Setup: RL with policy gradients The basic setup is that we want to optimize…
Modified: July 06, 2022.
Language is an incredible bottleneck. There are infinitely many true facts about the world, even just in pure math, and yet we communicate…
Modified: January 23, 2022.
The reason to try new things is not really because the new things themselves are more exciting than the old ones. The reason is that it…
Modified: June 24, 2020.
Modified: .
Modified: May 14, 2021.
From Jeff Bezos' 1997 shareholder letter : Some decisions are consequential and irreversible or nearly irreversible – one-way doors – and…
Modified: January 18, 2021.
Inspired by Kevin Buzzard's overview of the state of automatic theorem provers. Type theory is like set theory in that sets and types are…
Modified: December 23, 2021.
nostalgebraist argues that unconditional love can't and shouldn't exist : A parent might love their child "unconditionally," in the well…
Modified: April 12, 2023.
update April 2024: I'm going to leave this here, but I now think about confidence in less of an information-theoretic belief way, and more…
Modified: April 16, 2024.
It's a basic law of probability that, given two events A and B, the probability that at least one of them occurs is given by This counts the…
Modified: March 02, 2022.
(this note expresses a tendency that I notice in myself. I don't necessarily endorse this tendency but I think it's interesting to…
Modified: July 07, 2023.
Modified: .
Modified: .
From a review by [ Oliver Burkeman ] of Jordan Peterson's "Beyond Order" ( https://www.theguardian.com/books/2021/mar/02/beyond-order-by…
Modified: February 22, 2022.
in contrast to [ things I believe that no one else believes ], which are intended to be potentially-novel insights about the world…
Modified: March 25, 2024.
Modified: .
like a 'useful perpective', but 'lens' implies focus or distortion whereas 'perspective' implies linear projection. Related to [ many models…
Modified: January 17, 2021.
Consuming unstructured content from the internet is addictive. Twitter is full of life advice, interesting technical discussion, takes on…
Modified: May 29, 2020.
Modified: February 10, 2022.
Suppose I have an agent that generates text. I want it to generate text that is [ value alignment|aligned ] with human values. Approaches…
Modified: February 21, 2022.
Modified: February 21, 2022.
Modified: September 11, 2020.
Notes on the Alignment Forum's Value Learning sequence curated by Rohin Shah. ambitious value learning : the idea of learning 'the human…
Modified: April 07, 2023.
The standard [ Markov decision process ] formalism includes a reward function ; the total (discounted) reward across a trajectory is its…
Modified: October 16, 2022.
References: Jacob Eisner, High-Level Explanation of Variational Inference (2011) https://www.cs.jhu.edu/~jason/tutorials/variational.html…
Modified: April 26, 2022.
Holy shit. In December on Galiano I was brainstorming about [ continuous structure learning ] and thought of the general trick, for…
Modified: June 09, 2020.
The divergence of a vector-valued function on a vector field measures the extent to which a given point is a source of the field. It…
Modified: June 08, 2024.
Inspired by [ Emily ], I'm considering going 'mostly' vegetarian. What would that mean for me? I don't myself buy meats or dairy products…
Modified: March 07, 2020.
I don't know quite how to articulate or formalize this, but I get a sense that there is something fundamentally analogue, 'periodic' or…
Modified: March 19, 2024.
Why am I doing all of this? If I carve aside hours or days or months to 'fill in' my graph of notes, what am I hoping to get from it? Why is…
Modified: February 25, 2022.
Ref: https://arxiv.org/abs/2010.11929 We start by chunking an image into patches, and concatenating each patch with a position embedding…
Modified: .
Telling people about your failures, your fears, your self-doubt, your insecurities can be a path towards deeper connection. Understanding…
Modified: September 21, 2021.
Modified: .
How to be warm: https://www.youtube.com/watch?v=1MolmoFuXu4&t=123s
Modified: November 27, 2023.
The "strength of weak ties": most good things in life come from people you barely know. This is because your close, regular connections are…
Modified: April 26, 2024.
Thinking through: Why the toughest capitalists should root for a wealth tax ( https://www.ft.com/content/e1adf707-b95a-4422-9211-1841cd7ce…
Modified: May 09, 2021.
Moxie Marlinspike on web3: https://moxie.org/2022/01/07/web3-first-impressions.html We know that people do not want to run their own…
Modified: January 07, 2022.
[ weekly review ] • Plus: What went well? • Minus: What didn't go so well? • Next: What will I focus on next week?
Modified: January 23, 2022.
Reference: Mahmood et al., 2014. Weighted importance sampling for off-policy learning with linear function approximation Here's a situation…
Modified: April 23, 2022.
I suspect many of these are evergreen. I'm not [ writing ] enough. I'm not keeping up a regular journaling practice.
Modified: .
Just like norms in the Trump administration, there are mental habits, rhythms of life, attitudes towards the world, that are powerfully…
Modified: February 23, 2020.
In the course of any person's life, you take in a vast amount of information. You have your own personal experiences, of course, and you…
Modified: June 12, 2021.
See also: [ if ever a prof ], [ advice for college students ] Things not directly related to course material that I wish I'd learned earlier…
Modified: August 28, 2021.
What will I do when I don't have a job? I don't feel that I have a clear direction. I want to learn and explore. There are lots of [ my…
Modified: March 04, 2022.
A story from [ Dan Brown ]: A group of psychologists came to interview the Dalai Lama, the spiritual leader of Tibet. One of the Americans…
Modified: February 10, 2022.
As a kid, we learned about https://en.wikipedia.org/wiki/The_Game_(mind_game) : if you think of the game, you lose. (and have to say "I…
Modified: January 15, 2022.
From 2017: wisdom I've acquired: the psychology of depression. :-( and grad school. :-( and being gay. [ dual-process cognition ] theory…
Modified: February 15, 2022.
“It was true that I didn’t have much ambition, but there ought to be a place for people without ambition, I mean a better place than the one…
Modified: February 10, 2022.
Modified: .
This may be a central point of confusion: how do we define AI systems that have preferences about the real world , so that their goals and…
Modified: April 12, 2023.
In software: a library is a collection of tools. You can use some or all of them, in combination with other tools. A framework , on the…
Modified: May 08, 2020.
your writing needs to be at the edge of your knowledge, it needs to address the most fascinating people you know or can imagine. That is…
Modified: October 03, 2023.
Quote I like from Manuel Blum's advice to grad students , connecting writing to the power of [ Turing machine ]s: STUDYING: You are all…
Modified: September 13, 2022.
What is the philosophy of the project? What principles is it betting on? Example from Ben's Ads doc: iterating on an end-to-end pipeline…
Modified: June 25, 2022.
Regular writing practices that would be valuable. [ prediction as a model-building exercise ]
Modified: February 14, 2021.
Modified: May 01, 2020.
The models we use in AI are [ all models are wrong|wrong ] (if maybe still useful). How? Agency The [ agent ] model assumes a separation of…
Modified: February 13, 2022.
status: a theory that feels true for my personal trajectory. Totally uncritiqued and unverified that anyone else shares this experience…
Modified: May 22, 2021.
yin is being yang is doing there is a profound relationship between those two at a deep level and/but there is a whole web of associations…
Modified: March 26, 2025.
Modified: July 10, 2020.
Sometimes it's daunting how much knowledge there is in the world. For any given area, there are a thousand specialties and subspecialties…
Modified: April 17, 2022.
Something that SuccessfulFriend said today: It's rare that someone totally independent comes up with a really good idea. The best ideas come…
Modified: February 25, 2022.
A zero-knowledge proof allows a prover to demonstrate that it possesses certain information, without revealing that information to the…
Modified: October 23, 2022.
A zk-SNARK, or zero knowledge Succinct Non-interactive Argument of Knowledge, is a [ zero knowledge ] proof system that is non-interactive…
Modified: October 23, 2022.