Info Diet gives a glimpse into the personally chosen feeds and media routines of the people building what comes next. In every edition, one different builder spends two days logging everything they take in to keep ahead of the curve. This time: Elena Burger, a deal partner on the a16z crypto team, where she concentrates on games, NFTs, web3 media, and infrastructure. Elena lives in New York City.
Sunday, July 10
10 a.m.: This is a strangely slow Sunday for me — I usually wake up by 7 or 8 a.m., drink coffee, head out for a run in Central Park, and check Twitter and my email. But today I’m staying at a family member’s beach house with a friend, so my usual morning energy has been dialed down a bit.
I drink coffee at the kitchen table and open a copy of The New Yorker in front of me. That’s unusual too: I used to consume a lot of print media, but these days most of what I read is digital. I skim a profile of Mavis Staples, the gospel/blues singer, while people drift into the breakfast room.
12 p.m.: I’m riding the Jitney bus back to Manhattan when my friend puts one of his airpods in my ear and asks, “Have you ever watched Channel 5?” No, I hadn’t.
Channel 5, I learn, is Andrew Callaghan’s man-on-the-street interview series on YouTube, where he goes after potentially contentious settings (a pickup artist bootcamp, an anti-vax rally) and records the people involved doing their thing. We watch the Utah Rap Festival episode (I’d suggest turning your volume down to a little under “max” before you click), which includes some genuinely impressive rapping from a 15-year-old. I like the video — it’s somewhat reminiscent of shows like Nathan For You or How To with John Wilson. I don’t spend much time on YouTube, but like the Mavis Staples profile, if something is right in front of me, I usually take it in.
I then put in my airpods and listen to the Lex Fridman interview with Grimes from a few weeks back. I come away from it completely loving her. I hadn’t really grasped how far into crypto she has gone or that she holds a very techno-utopian view of web3. She talks about things like using DAOs to support motherhood and open-source research, and also about how web3 or AI can automate the role of music managers and other intermediaries. It’s so fascinating to me that artists can immediately understand crypto use cases far more practically than many people in finance or traditional tech can. I’m also now very curious about what Mavis Staples would think about DAOs.
On a separate note, there’s a point in the podcast where Lex says he lives next door to a famous anarchist, which makes me ask: if you’re an anarchist and you do own private property, what principles of anarchism do you really still bother to follow? If anyone knows, please ping me.
3:30 p.m.: I’m back home. I open the SITALWeek (Stuff I Thought About Last Week) newsletter on my laptop, which I read every Sunday. Brad Slingerlend puts the newsletter together, and he covers a lot of traditional finance and tech subjects: AI, semiconductors, the streaming economy, inflation, and oil/natural energy. Since I spend 95% of my time thinking about crypto, it’s a great “touch grass” equalizer for me.
This issue includes a strong piece on how robotic surgery affects the skill levels of trainee medical students, China’s lead in the lithium-ion battery market, and a few intriguing links about approaching austerity spending across various streaming services.
Some of these links, especially the ones about streaming/online media, really do seem crypto-adjacent — and even though this is my “non-crypto” newsletter, I can’t stop thinking about ways web3 might step in. It feels like we may be hitting the ceiling of what big streamers can efficiently and accurately model when it comes to user preference and stickiness, and newer crypto business models can let more artists keep a bigger share of the value they generate while also letting the community participate in IP creation. (Not to self-promote, but I wrote a Twitter thread on this here).
4:30 p.m.: I head out for a walk and call my brother. He’s an electrical engineer, and for the last few months I’ve been trying different ways to crypto-pill him. I’ve probably been a little too forceful lately because the first thing he says when he picks up is “Elena, I’m not coming to the Stanford Blockchain Conference.” Alas.
We start talking about the recent leaps in AI and image generation, and we discuss DALL-E, which I recently got access to and have been having lots of fun with. My brother sends me an Astral Codex Ten piece called “A Guide To Asking Robots To Design Stained Glass Windows,” where the writer explains his method of steering DALL-E asymptotically closer to the picture he wants produced.
When I use DALL-E, I switch back and forth between trying to overpower it with highly exact prompts and intentionally giving it something loose and somewhat vague. For example, a few weeks ago, I asked it for “Thomas Edison selling bootleg DVDs on the street in the 90’s.” The first two images it gave me look like they’re from the 1990s (saturated colors, Tom-Ed beside a pickup truck), but the other four look like they’re from the 1890s — black-and-white, grainy images, and the DVDs are absurdly huge and oblong, like the only way to watch them would be to run them through a visual phonograph. This was a creative jump I don’t think I would have been able to make, so it’s fun to imagine that DALL-E arrived there by accident (or on purpose…).
After reading the Astral Codex Ten piece, I realize I can sort of answer the question I had earlier (“What would Mavis Staples think about DAOs?”) if I put it through GPT-3, which I also can access through the OpenAI playground. I enter the prompt: Write a paragraph from a New Yorker article about Mavis Staples talking about cryptocurrencies and decentralized autonomous organizations.
Here’s the result:
Mavis Staples has been singing gospel and soul music since the nineteen-fifties, and she’s as admired as a living legend as she is loved as a performer. This past fall, she released her latest album, “If All I Was Was Black,” a collaboration with the producer and musician Jeff Tweedy. Staples is also a venture capitalist. In early December, she took part in a panel at the Web Summit conference, in Lisbon, on the topic of cryptocurrencies and decentralized autonomous organizations. “I’m not a techie, but I’m learning,” she said.
Pretty amazing!! The only mistake I spot (aside from the made-up details about Mavis Staples’ second life as a VC showing up at Lisbon crypto conferences) is that Mavis’ most recent album is “We Get By,” which came out in 2019. It’s entirely possible that this is because the training data GPT-3 used isn’t fully current, though I’m not sure.
7:30 p.m.: I don’t really remember how I wandered into this rabbit hole, but I’m reading about Minecraft servers. I come across this excellent 2015 Rock Paper Shotgun piece on the rise of Hypixel, the largest independent server by CCU (concurrent users). It reached a peak load of 216k players in 2021, and typically averages about 45-70k players/day. I’m still trying to work out how Minetrack measures CCU’s in the context of Hypixel games, because the server itself funnels players into much smaller “parties” where the games are played (so I don’t think you’re actually seeing tens of thousands of players rendered around you). But I could be mistaken. Scaling games to tens of thousands or even millions of CCU’s is still a huge engineering challenge — Water & Music recently published a very good writeup on this in its “9 design principles for a musical metaverse” (skip to the “Scaled Synchrony” section for more on this).
Monday, July 11
7 a.m.: I get up and do a few laps around Central Park. I only listen to music when I run (I tried podcasts for awhile, but my mind drifts when I jog), and right now I’m playing a lot of Meek Mill, artists like SOPHIE, and aughties-era producers like Ryan Leslie.
10 a.m.: Once again, I’m on my “non-crypto” news circuit. At this point even non-crypto news is crypto related: chip shortages, inflation, entertainment, etc — it’s all connected. One of my other regular newsletters is Liberty’s Highlights, which is a great roundup of news about the more earthly parts of the world: semiconductors, chips, traditional media, etc. This issue actually references yesterday’s SITALWeek link about surgical automation — there are also some interesting links from The Wall Street Journal and Compound Capital Advisors CEO Charlie Bilello, respectively, about the energy efficiency of electric vehicle batteries and Eurozone inflation.
8 p.m.: Most of my day gets eaten by calls or answering emails and Slack messages — not much time to consume news (other than tweets). I do read a lot of pitch decks and emails from founders, but my time for closely reading articles or falling down rabbit holes is saved for later at night (or on weekends).
Tonight, I unwind by catching up on the Reddit AMA with the Ethereum Foundation research team from a week ago. I definitely try to stay as current as possible on conversations coming out of the EF (especially as we approach the Merge). It’s pretty interesting to read through the AMA and see that the core devs don’t always fully agree (a good example is the two answers to this question about the gap between research and implementation).
The biggest area of consensus seems to be around the kinds of applications that people are looking forward to in the post-rollup, scaled-Ethereum future: the agreement (among Vitalik, Justin Drake and Danny Ryan, at least) is that applications around identity, reputation, and privacy will see a lot of innovation.
(Another newsletter I keep returning to is What’s New in Eth2, put together by Ben Edgington at ConsenSys. And while this is on my mind, Tim Beiko deserves a special mention for the remarkably thorough notes he shares from the recurring Ethereum AllCoreDevs calls.)
1 a.m.: After dinner, I continue to catch up on messages and scroll through Twitter. Then, I’m somehow back in a Minecraft rabbit hole. Before drifting off to sleep, I read this OpenAI research post on training a neural net to find and craft items in the game. Impressively, only a small portion of the videos used to train the neural net were labeled with corresponding mouse and keyboard actions. The rest was just an enormous feed of videos uploaded by players online (in other words, the neural network was able to extrapolate and learn from a small set of labeled data, and apply it to a much larger set of unlabeled data).
The more I pay attention to what is happening in AI, the more I keep wondering how the field will meet crypto. Will we see more protocols that try to build on-chain AI as a community-run tool that anyone can inspect and improve? 0xPARC presented a proof-of-concept a few months ago showing an on-chain numerical image classifier inside a zkSNARK, and a few years ago Ali on our team wrote a strong Twitter thread on how one might bootstrap an autonomous neural network on-chain. I hope we keep seeing more applications like these in the future.
***
Get Elena’s Info Diet
Read
- “The Gospel According to Mavis Staples” by David Remnick [The New Yorker] “A Guide to Asking Robots to Design Stained Glass Windows“ [Astral Codex Ten] “The Future of Minecraft’s Biggest Servers” by Alex Wiltshire [Rock Paper Shotgun] “9 Design Principles for a Musical Metaverse” [Water & Music] “[AMA] We Are EF Research” [Reddit] “Learning to Play Minecraft with Video PreTraining (VPT)” [Open AI] “ZK Machine Learning” [0xPARC]
Listen
- “Grimes: Music, AI, and the Future of Humanity” [Lex Fridman Podcast]
Watch
- “Utah Rap Festival” [Channel 5 with Andrew Callaghan]
Subscribe
- SITALWeek [NZS Capital] Liberty’s Highlights What’s New in Eth2 [Ben Edgington]
Follow
- @benjaminion_xyz, @TimBeiko and @alive_eth [Twitter]