Links (5)
And all who heard should see them there, And all should cry, Beware! Beware! His flashing eyes, his floating hair!
Disclaimer: Just because I include a link does not mean I endorse it or am confident in its claims. That said, to date I have not received a single correction from any of you readers, so that must mean every single link I’ve shared so far has been 100% correct. This is how the internet works, right?
Addendum: The original version of this post that was sent over email had a few links with “[Add description]” placeholders rather than blurbs. I have fixed the web version, but unfortunately cannot take back the email and edit. If you noticed these incomplete blurbs and experienced any frustration as a result, I apologize!
AI & Technology
My response to AI 2027: Vitalik Buterin responds to AI 2027, arguing that defense may not, or at least doesn’t have to, lag so far behind offensive AI capabilities if we are in a takeoff.
Superhuman AI in a Normal Age by Zhengdong Wang: Fictional story riffing on the synthesis of AI 2027 and AI as Normal Technology.
And so it was that the hedgehog and the fox alike became philosophers in the truest sense of the word, and lived happily ever after.
The Outsize Impact of AI Logistics by Austin Vernon: Austin on how big a deal AI logistics could be. My general sentiment on Austin’s breadth and depth continues to be: “How does he do it?”
Some Thoughts on Medical Superintelligence by Ankit Gupta: Thoughts on the implication of Microsoft’s recent medical AI work and its implications for startups.
The American DeepSeek Project by Nathan Lambert: Nathan Lambert, of Interconnects fame, with a vision for American open (lower-case “o”) AI. I support giving Nathan $.
Rlver: Reinforcement Learning With Verifiable Emotion Rewards for Empathetic Agents by Peisong Wang et al.: We are about to hit levels of glazing beyond all comprehension.
Why Aren't LLMs General Intelligence Yet? by John David Pressman: Interesting thoughts on what’s missing from current models by someone who’s been doing some of the more interesting work on agents.
This is an academic convenience, in reality rewards do not come from "the environment" they're identified by the agent through sensory organs, feature extraction, symbol binding, etc. So for your agent to be generalist in the sense that it can be put in front of an Atari 2600 and start playing without a human laying out what actions are worth what for it, it needs to be able to autonomously formulate, valuate, and verify the completion of goals. It needs to be able to write and execute reward programs without wireheading (i.e. always giving itself the maximal score). My Weave-Agent and Zhao et al's Absolute Zero framework both do this and scant little else I'm aware of so this may very well be the bottleneck (since I ran out of compute before I could confirm that it wasn't).
The road to Top 1: How XBOW did it: XBOW, an offense security AI company, built an agent that made it to the top of the HackerOne leaderboard. I think this is the first time I’ve seen a non-AI lab build an AI system which won a real world competition, especially outside of math or algorithmic coding. Beyond being an interesting case study, I’m excited to see d/acc ideas working in the wild.
Checking In on AI and the Big Five by Ben Thompson: Stratechery with a mid-season check in on how different AI players are doing.
Q-learning is not yet scalable by Seohong Park: Research-heavy blog post on why Q-learning, a workhorse RL algorithm, still doesn’t scale in the way self-supervised pretraining does.
Build for Freedom, Not Control by Cosmos Institute: A bit light on details but a good exhortation that progress's speed may be hard to change but its direction vector and shape may be more influenceable.
d/acc: one year later by Vitalik Buterin: More Vitalik, this time reflecting on d/acc one year since he originally wrote about it.
AutoBioML: Building a Generalist Multi-Agent Framework for BioML by H.M.: Good turpentine-y post on building a multi-agent system for comp. bio tasks.
Biotech & Medicine
Mapping the off-target effects of every FDA-approved drug in existence (EvE Bio) by Abhishaike Mahajan: Abhi on a very cool FRO.
Joining Noetik by Abhishaike Mahajan: Abhi laying out his thesis about promising AI / bio bets as part of announcing his next career adventure. (Congrats!)
What Are Virtual Cells? by Elliot Hershberg: Elliot back from his writing hiatus with a helpful post on virtual cells. If you've been following the space, much of this may be a rehash, but still serves as a pretty good reference/informal review.
Why Haven't Adenoviruses Caught up to AAVs for Gene Therapy? by Effie Klimi: FINALLY, someone who knows writes up the story behind why Adenoviruses aren’t used more for gene therapy. It’s not just beacuse of the Jesse Gelsinger tragedy.
How not to waste a billion dollars (on your clinical trial) — with Meri Beckwith by Development and Research: I'm really loving this new podcast on drug development and trials. So much good stuff from grounded but not jaded practitioners on how to approach stages besides discovery from an unconventional standpoint.
Drugs cheaper than beer — with Brian Finrow by Development and Research: See above.
Induced Phosphorylation by Vera Mucaj: Really good deep dive on an emerging “frontier in drugability”. Phosphorylation can be thought of as the substrate for much of the cell’s protein logic, so being able to modulate it precisely could be really powerful.
A Visual Guide to Genome Editors by Evan DeTurk: Amazing summary of the molecular biology and function of different editors. This is one of those summaries that I can easily imagine catalyzing progress by making it easier for new people to ramp up and even helping experienced people think clearly about the fundamental differences between different editors. I ended up making tens of Anki cards from this. I finally can now remember what the RuvC and HNH domains do!
Business & Strategy
Story behind the Windsurf acquisition: If you’ve been following the Windsurf saga and were wondering what happened behind the scenes, this interview fills in a lot of it from the Cognition perspective. Also, Scott Wu is really funny!
Bubble talk by Sam Altman: Scott Wu mentioned this in a clip of an interview I watched and so I went back and read it. TL;DR: 10 years ago, sama got tired of investors saying startups were a bubble and made some concrete predictions about startup valuations. At the time, his predictions presumably seemed aggressive, but in hindsight they were laughably tame.
"The top 6 US companies at http://fortune.com/2015/01/22/the-age-of-unicorns/ (Uber, Palantir, Airbnb, Dropbox, Pinterest, and SpaceX) are currently worth just over $100B. I am leaving out Snapchat because I couldn't get verification of its valuation. Proposition 1: On January 1st, 2020, these companies will be worth at least $200B in aggregate."
The Next Great Distribution Shift by Brian Balfour: One of the better articles on tech / AI strategy I've read in quite a while. Also credit for making what I feel is a pretty evaluatable-in-hindsight prediction. I'll certainly be tracking to see whether this holds true.
Solution-space taste by Grant Slatton: I'm probably the wrong audience for this because I'm pathologically attracted to simple, elegant solutions, but it's still the case that it distills something I often struggle to capture.
Amazon and the "profitless business model" fallacy by Eugene Wei: A good reminder how non-consensus Amazon was even post-IPO!
How Airbus took off by Alex Chalmers: If you’ve ever wondered how Airbus is crushing Boeing, this is the story for you. A detailed accounting of how Airbus was formed and its strategy.
Management & Organizations
Your Manager Is Not Your Best Friend by Stay SaaSy.
Politics & Society
American ideals matter in an age of rising authoritarianism by Peter Wildeford: Sorry, I really try to avoid (partisan) politics posting and I know this is another borderline one, but I enjoyed it too much to not include it.
When protesters in Hong Kong waved American flags while facing Chinese crackdown, they weren't endorsing every American policy. They were recognizing what the principles of 1776 represent in a world of narrowing possibilities. When Ukrainian soldiers carry flags signed by American supporters, they're acknowledging what the American experiment means for human freedom. Today, on July 4th, I ask every American to consider the same.
The signers of the Declaration of Independence didn’t know if their experiment would survive King George’s retribution. They signed anyway. Today we know democracy can work — we’ve seen it lift billions from poverty, spark the digital revolution, and create cultures that captivate the world. The question isn't whether America has failed to live up to its ideals — of course it has. The question is whether those ideals remain humanity’s best tool for building flourishing societies and whether those ideals are worth defending. Look at where refugees flee to, not from. Look at which values activists invoke when demanding better. Look at whose flag Hong Kong protesters wave.
The answer is obvious.
Happy Fourth of July.
What Can We Learn From Estonia? by Santi Ruiz: Another great Statecraft, this time on how Estonia built out such a dynamic, tech-first government.
"China's digging out of a crisis. And America's luck is wearing thin." — Ken Rogoff by Dwarkesh Patel:
The world of tomorrow by Virginia Postrel:
How exactly today’s longings might manifest themselves, whether in glamorous imagery or real-life social evolution, is hard to predict. But one thing is clear: For progress to be appealing, it must offer room for diverse pursuits and identities, permitting communities with different commitments and values to enjoy a landscape of pluralism without devolving into mutually hostile tribes. The ideal of the one best way passed long ago. It was glamorous in its day but glamour is an illusion.
Programming & Technical
The Art of Lisp & Writing by Richard Gabriel: I'm an unabashed early adopter of AI coding tools but I do wonder if, especially their more agentic variants, currently lose something that we had with languages like Python.
Miscellaneous
A former FBI agent traveled to Louisiana to ask a hired killer about a murder that haunted him by David Howard: Holy shit, this was so good, but also terrifying. A journalist and former FBI agent have a series of conversations with a now imprisoned, very successful hitman. The main story is about an unsolved murder, but for me the most interesting part was the insight into a real hitman’s psychology.
Was this sent too early? Some of the bullet points say [Add description]