Reading Period: January 01 - March 23
1. The Holloway Guide to Equity Compensation (P), by Joshua Levy
Link: https://www.goodreads.com/book/show/48753169-the-holloway-guide-to-equity-compensation
Simple book, very comprehensive overview of startup equity. I've been increasingly applying my financial expertise to the world of startups, and this was certainly a useful resource.
2. The Moral Circle (A), by Jeff Sebo
Link: https://www.goodreads.com/book/show/213395476-the-moral-circle
Jeff and I have a different set of arguments regarding digital minds. His are inherently probabilistic. If there is only a one in 1,000 chance of digital consciousness, isn't the scale big enough that this issue should be one of our top issues? This thinking is fair, and entirely rational, but I tend to avoid it in my own writings. Why? Well, I think it sort of anchors the audience into thinking that digital consciousness is unlikely. In reality, we have no idea, and there don't seem to be obvious reasons that consciousness must be a biological phenomenon.
Jeff also states that "morality is a marathon, not a sprint." I think this is certainly true, and thus it is unfortunate that we may have so little time before AGI. Which means we have to get moving! One last point on the book's content: Jeff states that how we treat silicon beings during our time in power may shape how they treat us during their time in power. For a few technical reasons I think this is incorrect, and somewhat of an unfortunate anthropomorphization. To expand on this, I mean that AI alignment is either going to work or it won't, and misaligned AI are unlikely to "punish" humans who treated digital minds poorly, any more than they are to "reward" humans who are emphatic towards digital minds. For this to happen would require creating a form of AI that to me seems unlikely, and if we can instill that sort of empathy in AI, have we not likely solved the alignment problem? Still, there may be some interesting ways in which this claim may be correct in a roundabout way. Perhaps focusing so clearly on empathy and treating AI development with the care that would be required to develop conscious beings (instead of calculators) is really the safest path forward for everyone.
Regardless, Jeff is an inspiration. I met him a few days before reading this book, and I would say that it is quite impressive that Jeff can stand in front of a crowd and talk about such "radical" ideas. His ability to press so far into caring about the "bots" not only keeps me going, it fueled me in the first place.
3. Venture Deals (P), by Brad Feld
Link: https://www.goodreads.com/book/show/11865558-venture-deals
Pretty incredible guidebook, filled to the brim with useful content. What matters in a term sheet? Economics and control. What's the deal with early-stage financial models? The only thing to know with certainty is that they will be wrong. Is a bridge loan a bridge to the next round? Or is it a pier that drops into the ocean? These questions, and a lot more, are all answered. The informal language is a massive benefit. For example, Brad states that "if you care about information rights for your shareholders, you are nuts. You should run a transparent organization as much as possible in the twenty-first century. If you can't commit to sending your shareholders a budget and financial statements, you shouldn't take on outside investors." Such clarity of communication is the books biggest strength, especially since the topic in general is so dense. Well worth the read, possibly multiple times if you work in the space.
4. Neural Networks and Deep Learning (P), by Michael Nielsen
Link: https://www.goodreads.com/book/show/24582662-neural-networks-and-deep-learning
Many instances of something seeming extremely complex, but are actually just explained by simple but powerful set of ideas. Michaels book guides us through all of these simple but powerful ideas. It is interesting to think through perceptron mathematics and gradient descent, but what I found most interesting was observing how my own mental model of neural networks changed during my reading of this book. It is hard to really start visualizing the process of backpropagation, but once you begin to, it becomes pretty strikingly straightforward. Probably the most interesting topic in the book came up late, when Michael discussed sort of the reason scale was so surprising. Occam's razor was at work in deep learning initially, when people were trying to stay small and create the most optimal solution. Turns out, more compute and more data simply make the models better, who knew? Also, the concept of universality is interesting. Basically, no matter what function we want to compute, there is a neural networks that can do the job. I'd recommend this book to anyone with even a slight interest in deep learning.
5. Langchain Crash Course (P), by Greg Lim
Link: https://www.goodreads.com/book/show/59713693-gpt-3
Not worth the read. I was interested in learning more about Langchain, but I would have been better served reading five minutes worth of online documentation instead.
6. GPT-3, by Shubham Kublik
Link: https://www.goodreads.com/book/show/59713693-gpt-3
Given how much time I've actually spent with LLMs, this book certainly wasn't worth the read. It was essentially a basic overview of how to use the OpenAI API, circa 2022.
7. Chess Story (P), by Stefan Zweig
Link: https://www.goodreads.com/book/show/59151.Chess_Story
Damn, this was shockingly good. Not many authors can cram a moving storyline, compelling characters, and a thought-provoking ending into a book that is less than one hundred pages. Stefan more than succeeded here. The ending battle of wits between Czentovic and Dr. B. displays such raw psychological intensity that it stands with some of the best fiction. Certainly worth the read.
8. The Mom Test (P), by Rob Fitzpatrick
Link: https://www.goodreads.com/book/show/52283963-the-mom-test
A fairly short and simple business book. The mom test: ask
questions to your customer in a roundabout way that avoids false positive (“oh
that’s a nice idea sweetie. I’d certainly buy a cookbook app that you made.”).
The book is filled with handy tips, such as: “People know what their problems
are, but they don’t know how to solve those problems” and “if they haven’t looked
for ways of solving it already, they’re not going to look for (or buy) yours.”
Probably only worth reading if you haven’t already read a ton of business
books.
9. Bluets (P), by Maggie Nelson
Link: https://www.goodreads.com/book/show/6798263-bluets
This book was certainty interesting. I was a bit thrown off by the random, sharp sexuality, but I don't believe there was much that was out of place in this book. It was an intriguing, experimental book that I'm sure many would enjoy. That being said, it was not for me. Through no fault of my own, and through no fault of Maggie, I simply didn't connect with this one. The last few sentences of the book are incredible, but I actually think it is worth otherwise skipping altogether. Here is the end:
"I want you to know, if you ever read this, there was a time when I would rather have had you by my side than any one of these words; I would rather have had you by my side than all the blue in the world. But now you are talking as if love were a consolation. Simone Weil warned otherwise. 'Love is not consolation,' she wrote. 'It is light.' All right then, let me try to rephrase. When I was alive, I aimed to be a student not of longing but of light."
10. Managing Oneself (A), by Peter Drucker
Link: https://www.goodreads.com/book/show/2477223.Managing_Oneself
I'd skip this, read something like High Output Management instead.
No comments:
Post a Comment