Holiday Reading
A collection of posts and books to tide us over until the New Year
As we noted in our last issue, we plan to have a bit of down time over the holidays, so we’re using this opportunity to suggest some reading material to keep you going over the break. Since many of our readers joined us fairly recently, we are resurfacing some of our old posts, and we’re also recommending some of our favorite reading from other writers.
As people who write a lot (more than average for computer scientists, I would guess) we also spend a lot of time thinking–and reading–about language. One of the best pieces I’ve ever read on language is from David Foster Wallace, and that inspired my take on the “On Premises vs On Premise” debate. As in much of my career, I managed to irritate plenty of people by taking a position between the two extremes, but mostly I wanted to inspire people to read the essay by Foster Wallace. He has a bit of a reputation (undeserved in my view) of being hard to read–perhaps due in part to his extensive reliance on footnotes–but I recommend that if you want to get a gentle introduction to his work, you could start at the essay noted above or “A Supposedly Fun Thing I'll Never Do Again”.
Larry has also written about his literary influences. These include his Hemingway reference a few weeks back, his glowing review of Tracy Kidder’s House, and his discussion of John McPhee’s “Draft No. 4: Reflections on The Writing Process”.
Given our interest in language, it should come as no surprise that we are intrigued by large language models. I first started studying AI in 1984–yes, really–and I recently happened to crack open the book by Patrick Winston from that year, titled simply “Artificial Intelligence”. The preface is amazing in that it would look perfectly timely in a book published today:
The field of Artificial Intelligence has changed enormously since the first edition of this book was published. Subjects in Artificial Intelligence are de rigueur for undergraduate computer-science majors, and stories on Artificial Intelligence are regularly featured in most of the reputable news magazines.
Ah yes, our brilliant AI future was just around the corner in 1984, just as it is today, and the media was all over it.
I think my piece on “looking inside LLMs” is the one I spent the most time researching this year, and that’s because the picture is complex. While the view of LLMs as “stochastic parrots” (or “spicy autocomplete”) is not far from my own, getting a closer look at how LLMs work internally gave me a better understanding of why there is so much confusion about what is possible, as well as optimism from many who work in the field. It is just unfortunate that there is so much hype and shallow thinking getting in the way of properly understanding the potential and current risks of these systems. We plan to produce a book about machine learning (as applied in the context of networking) next year, and it will be as hype-free as we can make it.
A lot of what we have written over the last two years has been about the increase in centralization of various aspects of the Internet. My own experience with SDN was one of leveraging the capabilities of modern distributed systems to create logically centralized abstractions as a way to overcome the limitations of fully decentralized network architectures. I talked about this with Nicira founder Martin Casado, and wrote about it in this post and in our SDN book. So I’m not entirely anti-centralization, but the tradeoffs are subtle.
It’s now more than a year since we moved to Mastodon, and that’s very much a story about the renaissance of decentralization in the social media sphere. Hardly a day goes by that we don’t feel good about the decision to leave the bird site. And in a reminder that there is more to decentralized social media than Mastodon, here is my talk on 60 Years of Networking (with a strong focus on decentralization) on peertube.
In terms of the articles from our early days that have seen the greatest interest, I would say that every time we talk about RPC vs TCP we get lots of traffic. So too with anything questioning the value of MPLS. The reaction to the last newsletter on Outrageous Opinions makes me think I might need to write something about the rise and fall of ATM.
What else should you read over the holidays? I enjoyed this article on Indexes from the New York Review of Books. For a literary magazine, the NYRB also does a nice job on covering issues of the Internet and society, as in this piece on the Internet and privacy. (The NYRB, by the way, is not related to other publications with New York in their title, and was founded in 1963 during a newspaper strike.)
We’re big fans of Cory Doctorow, whose rate of publication puts us to shame. His recent post on the AI bubble is a good example of his work, and he may well have given us 2023’s word of the year: “enshittification”. A good place to start on this topic is here.
My non-fiction reading for the holidays includes Chernobyl: History of a Tragedy, which I was inspired to pick up after watching the HBO series on Chernobyl. I thought I had a basic understanding of how nuclear reactors work, but I had massively underestimated just how complex they are. This book helps expose that complexity and the fatal combination of technical and human failures that followed.
Finally, I made a concerted effort to read more novels this year, and one book deserves mention here: The Private Memoirs and Confessions of a Justified Sinner. I think I can safely say that it’s the funniest 19th century novel I’ve read, and surprisingly timely. Also Scottish, which is how I came to hear about it. Thanks to Project Gutenberg–a project aligned with our goals as open source publishers–for making it available.
I had an exchange with Paul Francis about his work on distributed search after I quoted him in last week’s newsletter. My enthusiasm for his work was perhaps a bit misplaced, if I take Paul at face value:
Regarding Ingrid, distributed search could never work. First it would never be fast enough (or as fast as centralized alternatives). Even if you could solve this, it'd be hard as hell to deal with spam in a distributed system...Exploring the distributed alternative is the prerogative of the researcher, but it was a stupid idea.
Well, it was a good talk in any case.
Thanks for supporting us this year and we will return with fresh content in 2024.
Got an idea for something that Systems Approach should cover in either a book or a newsletter? Send us a note.
Follow us on Mastodon.