I shitpost about startups, fûtbol, and developer infrastructure.
San Francisco, CA
Joined February 2014
- Tweets2K
- Following1.4K
- Followers2.1K
- Likes11.7K
Pinned Tweet
Londonmaxxing Part 2! One of the greatest angel investors of all time and a good friend. There’s been a ton of building in the morning, now it’s time for football. Wild videos and photos in the thread. cc @mjasay (en absentia)
Not good
‼️ BREAKING: Go has issued an URGENT security update blocking 8 certificate authority keys over undisclosed security concerns.
It will reject certificate chains using these keys, potentially breaking connections for rebuilt applications.
The filter cannot be disabled.
Here's what we know: thehackernews.com/2026/10/go…
Help HBCUs and their students continue learning online and in-person. UNCF has an efficient operation. Donate today to help at bit.ly/38Gzgzw.
I donated in honor of William Mitchell III.
the source of my hope
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
Thanks @bot strange list in a couple places for sure.
I’m at BazelCon this year with a much bigger and stronger company than ever before. While only a subset of our customers are build engineers, they are oftentimes an unseen team doing a lot of the heavy lifting. Front and center in the agentic era. If you’re in the Dam, I see you!
Today we experienced the production product of a customer we had been working on for 18 months. It feels so gratifying that the hard work of our team is contributing to such an incredible technology.
Marcus Eagan retweeted
Vector databases are more than just an index! 👀
Booking(dot)com ran their own test with 100M embeddings, queries with filters and concurrent writes. 🔬
They concluded that purpose-built vector databases behave differently from “general-purpose search engines with vector capabilities added on.” 🚀
In Weaviate, the index and storage engine are built together. 🧬
HNSW is an in-memory graph persisted with an append-only log, while objects and filters live in a custom LSM store. Filters come out of the LSM store as bitmaps that steer the HNSW search (ACORN), and writes are cheap appends.
HFresh goes further: its vector clusters live directly in the LSM store, so the index rebalances itself in the background instead of needing full rebuilds. 💽
Other systems inherit storage built for something else. Page-based databases turn every update into new row versions, WAL and VACUUM work. Segment-based search engines build a separate graph per segment, so queries search every segment and merges rebuild graphs. 🏗️
The index matters. So does its integration with everything underneath it.
booking.ai/how-we-selected-t…
Ukraine procurement process can be < 12 hours.
That’s probably the fastest in the world.
Marcus Eagan retweeted
The Ford Model T in 1920 had 15,000 nodes in its supplier dependency graph before you hit dirt.
A modern BMW X5 has an order of magnitude more, with chains 6, 7, 8, 10 layers deep.
Automation expanded the supply chain. As building got easier, the bar for excellence went up, and you used suppliers to simplify the domain you had to master.
Software is going to do the same thing. Many more smaller companies will deliver excellence in niches.
Maybe fewer enormous VC-sized opportunities, though...