Mageek

Breadth vs. Depth.

Alg4Opt covers more topics, providing the motivation behind the algorithm, sometimes a basic derivation, and a concrete implementation. It has citations in the margin for more info.

Nocedal and Wright will go more in-depth on derivation, proving theorems, etc. Implementations are pseudocode, and fewer topics are covered.

To be clear, both are fiction, it’s just that one is fantasy and the other is set in modern reality.

The video doesn’t show how the car got into this situation

Rust has a method for enforcing better memory safety. That is great for deployed applications, but can be annoying when you’re still exploring / mutating your code to figure out the right shape of things.

The article says it: “We determined that due to the persistent orientation mismatch of the towed pickup truck and tow truck combination, the Waymo AV incorrectly predicted the future motion of the towed vehicle.” It was detected, but it predicted the truck would move in a way that it didn’t end up moving.

My mom was hit by a driver when she was biking and fell with her arm in front of the wheel. The driver then decided to pull forward and drove over her arm. So humans don’t really solve that problem.

NHTSA standing general order crash rates are a mandated and publicly available data source. That’s what the study is based on.

Another very good blog overview here (not mine): https://ianthehenry.com/posts/delaunay/

I love DCELs and have been tinkering with them in my own side projects lately. Very cool data structure!

Isn’t Waymo already operating a full service in Phoenix? Haven’t heard of many issues from there.

I highly recommend the course “Computer, Enhance!” by Casey Muratori on substack for those interested in how CPUs / assembly work. You get to decode byte code, simulate instructions, learn how the stack works, etc. It is really well-paced, gives you plenty of space to figure things out your own way (with reference material if you need it), and helped me get several “aha!” moments that solidified how things work.

I recently started the new class by Casey Muratori, “computer enhance”, that teaches performance-aware programming.

I also try to complete a small personal project every month and write about it.

“As of 2020, South Korea was the only country among the OECD members to have a rate below 1, giving it a shrinking population.”

Being below 2 gives you a shrinking population, because men are half the pop and don’t produce children directly.

You still run A*, it’s just on the nav mesh rather than on a grid.

The title is complete clickbait. It should say “no one’s short intuitive explanations of why planes stay in the air are fully correct”

A blog post in which the classic cart-pole swingup problem is solved with a hierarchical control method that breaks the problem down into a high-level search problem on top of low-level LQR controllers. Everything is derived and code is presented.

Chesterson’s Fence comes to mind. If you think you have an obvious solution, you’re almost surely missing something.

Killing people obviously has repercussions. Building more housing obviously has more complexities. Zoning is hard. Forcibly taking land is complicated. Building the other amenities (transportation, parking, etc.) is hard. Taxing billionaires is complicated. They fund other things, can put financial pressure in other places, etc. California’s exodus of the rich is a real problem. etc.

I think the real question of value here is “what am I missing such that I think obvious solution X solves the problem?”

the prompt wasn’t questioning why there aren’t more - it was asking where they hang out online.

My next door neighbor would scream at people passing by. I’d routinely be woken at 7am due to his swearing. Things started getting worse. He verbally threatened to kill people. He bashed in an RV parked on the road by his property, causing thousands in damage. He jumped off his balcony toward a neighbor he was threatening. I went, with some others, to the police. They couldn’t do anything. If he yelled on his propery, that was his business. If the RV owner wanted to press charges, that was their business. Oh, they could offer him help, but he’d have to take it. Shortly thereafter, I had heard him calmly yelling “You think I’m scared of the police?! I’m not scared of no *Ing police!” And why should he? They visited him multiple times a week and did nothing. Two weeks later he beat his mother to death with a tire iron. Yes, we need good ways to determine instability, but some of these cases are pretty clear cut, and we have a very real problem.

This is astounding! I’ve looked into papers on this topic before and found them indescipherable. I love the clean visualizations, the implementations, and the clear progression.

The Chinese room argument is claimed to show how consciousness cannot be based on computation. I totally disagree - it merely shows that as an external observer we cannot verify consciousness.

I absolutely love doing this. I often write in LaTeX, where new lines don’t effect the typeset output. It is so much easier to see git diffs, comment sentences out, move sentences, identify a sentence by its line number, etc. as well.

“Proofs from the Book” is a great (and IMO fun) book. The 3Blue1Brown YouTube channel is amazing as well.

I very much enjoyed this blog post! Thank you for the thorough overview

The interview process is supposed to measure (or attempt) a whole bunch of aspects. Coding / technical skills, but also communication skills, reasoning, logical assessment, context / intuition, social interaction, decision making under uncertainty, systems thinking, testing, etc. No serious interview process is going to rely exclusively on leetcode questions to fully vet anyone. It requires a balance of many things. Furthermore, good interviewers are aware of these multiple dimensions, and are evaluating candidates on any many levels as possible.

Yes, leetcode questions are a part of the process, and are often an early gate. However, one can perform “badly” but still score sufficiently well on the non-obvious metrics and make it through. Most interviewers don’t actually care if you have the standard libs memorized, and get that. You typically have a lot of leeway to demonstrate your knowledge, and staying open about what you’re thinking and doing goes a long way.

If you have sway in how your company interviews, consider being clear with candidates on all the dimensions they are being eval’d on, maybe let candidates choose take home assignments, ask candidates about projects they have worked on, etc. as part of the process.

The authors open-sourced their setup for Tufte-style books: https://github.com/sisl/tufte_algorithms_book

An HTM layer is one feedforward layer in a Heirarchical Temporal Memory network. See for example https://arxiv.org/abs/1511.00083 which trains one-layer HTM models. See for example https://timallanwheeler.com/blog/2022/01/09/hierarchical-tem... for a more approachable blog post about the same topic

An HTM layer requires far more parameters than a traditional deep neural network, on the order of gigabytes for one or two basic HTM layers. A cortical column is going to have many such layers, and a 1000-brains model will have many (thousands?) of cortical columns. In my opinion, the underlying idea is fantastic but the practical aspects of implementing it are daunting.

This is an area that a lot of people are probably interested in. We still don't really have practical robots, self-driving cars, and personal assistants, and some sort of general context-sensitive learning intelligence breakthrough is needed to get us there. Traditional machine learning and deep learning have not yet been sufficient to get us there, and brain-based approaches might. It would be good to hear about more of the companies pursuing this sort of thing (maybe with less of a focus on brains and more of a focus on general understanding), and how to proactively contribute beyond the numenta-specific content here. Its a start though!

This articles frames the Artemis program as an us vs. then contest with Space X. They cannot, and never could, compete with Space X. However, NASA and the U.S. government will certainly benefit a lot if Starship is successful. I’m sure most everyone, except maybe some higher ups, are rooting for Space X just the same as many civilians are.

Wow this is so clean and thorough. Love all of the diagrams