Knowledgebase

Links, references, and things worth knowing. Repos, articles, papers, and HN threads — each with a brief.

PAPER
Constitutional AI — Anthropic

The original Constitutional AI paper from Anthropic. Introduces the idea of using a set of principles (a "constitution") to guide AI behaviour via RLHF, reducing the need for human feedback on harmful outputs. The key insight — that you can use AI to critique AI output against a principle set — has had outsized influence on alignment research.

alignmentRLHFAnthropicsafety
REPO
htmx — High Power Tools for HTML

HTMX lets you access AJAX, CSS Transitions, WebSockets, and Server Sent Events directly in HTML using attributes. Interesting counter-movement to the JavaScript framework ecosystem — the bet is that the browser is already a capable application platform if you let it do HTTP properly. Worth understanding as a design constraint when building content-heavy sites.

webfrontendhypermedia
ARTICLE
The Bitter Lesson — Rich Sutton

Sutton's 2019 essay arguing that the biggest lesson from 70 years of AI research is that general methods leveraging computation outperform methods that leverage human knowledge. Scale wins, always. Essential reading for thinking about where hand-crafted techniques and domain knowledge actually add value vs. just slowing you down.

AIscalingRLfoundations