State of context engineering
Ever-larger context windows don’t necessarily lead to good results. You need to choose what LLMs see for any given task. The key question is: how do you architect context for maximum effectiveness? This presentation from Drew Breunig (who is editing an O’Reilly book on the subject) explains key patterns and gotchas.
Data structures, explained
It took the AI boom for business leaders to start caring about ontologies. But do they understand what the term means in this context? This post explains the difference between taxonomies, ontologies, knowledge graphs, semantic layers, and context graphs — key to understanding structured knowledge systems. In particular, good context requires well-designed ontologies.
The sleep framework
From Google research, a new framework for improving LLMs’ memories that is inspired by human sleep. That is, there are cycles, with downtime being used to consolidate and integrate what has been learned during more active times. A fascinating and promising structure for continuous learning.
Amazon on software development in the age of AI
A public memo from Amazon’s engineering leadership. Wisely, they’re acknowledging the change while embracing the long-term view. As Bezos suggested, the key isn’t to focus on what’s changing but what isn’t: deeply understanding customers and serving their needs.
Emacs turns 50
Speaking of the long-term: this is only tangentially related to AI and likely irrelevant to your business, but it’s been fifty years since the first version of my favorite text editor. How many other software applications are still actively used after half a century? (I’m typing these words into Emacs now.)