7 December 2025

14 links · Sunday

Titans + MIRAS: Helping AI have long-term memory

The Transformer architecture revolutionized sequence modeling with its introduction of attention, a mechanism by which models look back at earlier inputs to prioritize relevant input data. However, computational cost increases drastically with sequence length,…

Perl's decline was cultural

There's been a flurry of discussion on Hacker News and other tech forums about what killed Perl. I wrote a lot of Perl in the mid 90s and subsequently worked on some of the most trafficked sites on the web in mod_perl in the early 2000s, so I have some thought…

576 - Using LLMs at Oxide / RFD

LLM use varies widely, and the ramifications of those uses vary accordingly; it’s worth taking apart several of the (many) uses for LLMs.

Instantaneous Velocity Is a Trap

Personal website for Aaron Brethorst - Seattleite, technology leader, photographer, transit enthusiast, erstwhile non-runner.

Building a Dumber Team

Having kids is weird. Here are these human beings, with their own interests, likes and dislikes, and absolutely no knowledge. I mean yes, over time they build up experiences, but they start out wit…

@tomlarkworthy/unaggregating-cloudwatch-metrics

Visnu Pitiyanuvath of Observable advocates "never aggregate your datapoints". The outliers are the signal! His talk made me realize that we, the developer community, could do better, especially the Cloud providers.