A note to readers: the proper care and feeding of a now-12-month-old has taken its toll on my writing. Here, I present a timeboxed kernel (or a draft,

Dave Guarino's Occasional Newsletter

submited by

Style Pass

2024-11-27 05:00:14

A note to readers: the proper care and feeding of a now-12-month-old has taken its toll on my writing. Here, I present a timeboxed kernel (or a draft, or a sketch) of something I’ve been thinking about. It is a seed of something likely to end up on Propel’s Insights blog, as it reflects ongoing research into the use of AI in benefits navigation. My hope is this strategy yields more regular output! As always I am grateful for thoughtful comments and emails in reply.

Applying AI to benefits navigation, I think I’ve seen a shape or pattern I’d like to highlight. I doubt it’s novel, but I do think it’s underlooked.

Specifically, when you start playing around with the frontier large language models (Claude, ChatGPT, Gemini) and you try real cases of problems users have navigating a program like SNAP, I think you find yourself, well, a bit epistemically cornered?

This is knowable with reference to specific policy documents. In fact, I can point you to the document updating income limits (and other numbers) for the 2024 COLA adjustment on the USDA-FNS website.