Travis Lake Johnson

I have never seen people as uniformly unhappy about a technology as they are about AI agents. I imagine this is what the Manhattan Project scientists must have been like after Trinity?

ugh

Just did this with a couple of friends irl. Feels like I've properly read a book for the first time in my life. We got a much better sense of how all the motivating questions and pieces of evidence actually fit together in the thesis. Asking each other very basic questions (and then trying to answer them) lead us to realize how murky our map of the terrain really was. And how confused our original interpretation of seemingly simple concepts was.

We've internalized that college costs a lot. But tuition actually doesn't. What costs a lot is housing. Look at UC Berkeley's breakdown. Tuition is under $18K/year, which is far less than any private high school in the Bay Area. On-campus housing meanwhile is over $22K. 🧵

If you think you can create better jobs numbers than the BLS then great, probably harder than you think but if you can, you can make a shitload doing macro trading and absolutely nothing is stopping you.

MASSIVE claim in this paper. AI Architectural breakthroughs can be scaled computationally, transforming research progress from a human-limited to a computation-scalable process. So it turns architecture discovery into a compute‑bound process, opening a path to self‑accelerating model evolution without waiting for human intuition. The paper shows that an all‑AI research loop can invent novel model architectures faster than humans, and the authors prove it by uncovering 106 record‑setting linear‑attention designs that outshine human baselines. Right now, most architecture search tools only fine‑tune blocks that people already proposed, so progress crawls at the pace of human trial‑and‑error. 🧩 Why we needed a fresh approach Human researchers tire quickly, and their search space is narrow. As model families multiply, deciding which tweak matters becomes guesswork, so whole research agendas stall while hardware idles. 🤖 Meet ASI‑ARCH, the self‑driving lab The team wired together three LLM‑based roles. A “Researcher” dreams up code, an “Engineer” trains and debugs it, and an “Analyst” mines the results for patterns, feeding insights back to the next round. A memory store keeps every motivation, code diff, and metric so the agents never repeat themselves. 📈 Across 1,773 experiments and 20,000 GPU hours, a straight line emerged between compute spent and new SOTA hits. Add hardware, and the system keeps finding winners without extra coffee or conferences.

By the way, climate change is a good analogy, here. The fertility crises poses at least as much of a threat to the future of humanity as climate change (and I believe in climate change). If there was as much cultural/media/elite/societal energy spent on raising awareness of the threat, people would come up with policies. Germany essentially sacrificed its whole energy grid to the gods of climate change.

Expectations now down to $7 billion.