Info Diet gives a glimpse into the personally chosen feeds and media routines of the people building what comes next. In each edition, a different builder spends two days logging everything they read, follow, listen to, and watch to stay ahead of the future. This week: Amjad Masad, cofounder and CEO of the online coding platform Replit. A software engineer by training, Amjad was born and raised in Jordan and now lives in Palo Alto, CA. His mission is to make programming easier to access: “I think it’s absurd that while software is eating the world, very few people know how to make it.”
***
Saturday
8 a.m.: I get up, make my standard triple-shot espresso, and check my phone to make sure nothing is on fire. (All good.) I open Pocket, my read-later app, and browse my saved articles. A few are about AlphaCode, DeepMind’s new AI system that generates code at a competitive level. From what I’ve read before, it has made DeepMind seem as if it may have cracked artificial general intelligence — which is impossible. That would imply the end of the world as we know it. I decide I should go straight to the source, so I open DeepMind’s 73-page paper and read the most pertinent sections. The findings are striking, but there’s still a brute-force quality to their setup that makes me question whether the AI is really “thinking” through the problem.
3 p.m.: I return home from a gym-and-sauna workout and I’m ravenous. (I’ve been fasting since yesterday.) I air-fry a burger for myself and scroll Twitter as I eat. I come across a conversation about web3, as one does these days, and an article someone posted called “When the Stagnation Goes Virtual.” It appears on Palladium, a site I’d never heard of before. The piece is very well written; it uses the metaverse-and-NFT craze to build toward a broader critique of our generation. Its argument is that we’re turning away from the physical world in favor of the virtual, which is no different from earlier generations seeking escape through drugs. I’m not sure I buy the author’s conclusion, but it does get me thinking.
I click through the rest of the site and I’m impressed, so I sign up for Palladium’s newsletter. Two other newsletters I’ve been liking are Astral Codex Ten, a psychiatrist’s reflections on reasoning, philosophy, science, AI, and more; and The Diff, which covers tech and finance.
9 p.m.: After dinner, I go to our home office to read some more. As my morning reading probably suggests, I’ve been interested in (OK, maybe even a little obsessed with) AI lately. I find transformer-based language models fascinating, especially when they’re used for code. This is something we’re beginning to use at Replit: We integrated an AI that explains a piece of code in plain English.
I’ve learned to trust my gut when something catches my attention. With that in mind, I picked up the classic AI textbook, Artificial Intelligence: A Modern Approach. But it was too bulky. I realized I needed something more hands-on and concise. I’ve heard strong things about fast.ai, an online video course centered on the hugely popular machine-learning Python library PyTorch. I start the first lesson. It lasts about an hour. Partway through, I get pulled into a group chat with my friends. Group chats have been a great way to talk about current events in a safe space, away from the mobs on social media. Then, back to fast.ai. I finish the lesson and call it a night.
Monday
8 a.m.: Once I’m out of bed and holding my espresso, I try to read something on my phone. But my son comes running toward me, wanting to play. I still squeeze in some Twitter scrolling before my 9 a.m. meeting. Nothing very interesting today.
I like Twitter lists. I called my favorite one “High Signal.” One person on it is Paul Graham, the founder of Y Combinator (and a good friend), who always brings fresh, noteworthy ideas. Another is Shane Parrish, the podcaster and blogger behind Farnam Street.
9 p.m.: I had nonstop meetings and very little downtime today. No media consumed, unless Zooms and emails count. But now I’m finished for the night, so I settle in and read by the fireplace.
Lately, in addition to AI, I’ve also been fascinated by the mind. How do our brains function? Where does consciousness arise from? Right now, I’m reading Surfing Uncertainty by philosopher Andy Clark, a book that sits right at the crossroads of my interests. In it, Clark draws on modern machine learning research to explain different aspects of cognition. Clark’s central thesis is that the brain does just one thing: prediction. (Incidentally, this is what artificial neural networks do as well.) So his argument is that consciousness is an emergent property of the brain trying to “predict” the sensory input it receives. It’s a cool, convincing theory. But I can only make it through a couple of pages before my mind is fried. I’m ready to unwind and not do any deep thinking.
10:30 p.m.: I mindlessly scroll through Twitter while I wind down. I try to go to bed at the exact same time every night. I’ve found that’s the best way to get a good night’s sleep. Once again, nothing very interesting, so I go to YouTube. I watch a couple of Joe Rogan clips on Powerful JRE, a channel that collects highlights from the podcast.
When the clock hits 11, I put my phone down and (try to) switch off my mind.
***
Get Amjad’s Info Diet
Read:
- “Competition-Level Code Generation with AlphaCode,” by DeepMind “When the Stagnation Goes Virtual,” by Ginevra Davis (Palladium) Astral Codex Ten, by Dr. Scott Alexander (Substack) The Diff, by Byrne Hobart and Jack Wiseman (Substack) Surfing Uncertainty, by Andy Clark
Watch:
- Powerful JRE (Youtube)
Follow:
- Paul Graham Shane Parrish
Learn:
- Fast.ai
Arrange: