I wanted to know whether embedding spreadsheet cells recovers what a column is for. The table understanding literature has been answering this for seven years, and the answer is mostly no, with a useful exception.
August 2, 2026
Notes on security, agents, and research.
I wanted to know whether embedding spreadsheet cells recovers what a column is for. The table understanding literature has been answering this for seven years, and the answer is mostly no, with a useful exception.
August 2, 2026
eBPF tooling like Tetragon is excellent at seeing what a workload does. For sandboxing an agent, seeing is the wrong verb. Here is why the enforcement point has to sit in the syscall path itself.
July 5, 2026
Agent frameworks catch prompt injection at each hop. Do the arithmetic across five hops and the end-to-end detection rate collapses to about 17%. The problem is structural, not a tuning issue.
June 13, 2026
Class-level code summarization is bottlenecked by data quality, not model capacity. Filtering with a single similarity score does not fix it. Here is the case for treating dataset selection as a search problem.
April 16, 2026
I moved to Lisbon for a while and learned the language on the way. The language was not the point. The people were.
July 22, 2025
We fine-tuned code models two ways and the cheap method won on quality, not just cost. Here is what that means for anyone choosing a training strategy.
May 8, 2025
I spent a while assuming that being good at the work was the hard part of getting the work. It is not, and figuring out why changed how I approach every conversation.
November 6, 2024
We scanned the Tranco Top 1 Million for service discovery records. Around 91% of domains advertising a service do it without working DNSSEC, and a lot of what they advertise is unencrypted.
August 14, 2024
I travel alone more than most people think is sensible. The reason is not the places. It's what happens to you when nobody around you knows who you are.
September 14, 2023