geddes munson✓

Love what @saalevine is doing here. It takes real and hard engineering work to make support bots BETTER than phone trees and call centers, with faster outcomes and higher csat, but that should be the standard. Lazy bots that do nothing but deflect give them all a bad name.

The top comment [1] on the GLM-Legal thread on HN gets into 'how will juniors learn legal things when the task that traditionally they did to learn are automated away' - the same problem we have been grappling with in tech. One of the answers is intentionality - as Claire, who just finished her new grad year at Affirm, writes up in this very real post: [1]

RT @allanca: @levie And that makes it different from the bubble in 2000. The demand is wayyyy ahead of the computer supply already and there appears to be no catching up. In 2000, we were building out fiber hoping the demand would come (and it did. Eventually).

I often find myself pushing hiring managers to do more reference calls and to do them better. Lots of great advice on doing it right from @brettberson > At the end of every single reference call, I ask the reference for the names of other people who would have a perspective on the candidate and use that to expand my list of potential people to call.

Over the years, I’ve become obsessed with references, so I decided to sit down and write out some of what I’ve figured out.

Our tutorial on Tabular Foundation models was accepted to the NeurIPS 2026 Education Track! Give it a try here to see how a pretrained model uses a table as context: Really great stuff coming out of our ML and ZT1 teams at @Affirm

RT @a_prout: We've done some work on HorizonDB to simplify PostgreSQL's interactions with cloud storage. Using storage designed for PostgreSQL lets PostgreSQL focus on running queries and transactions while letting the storage layer deal with everything else (i.e., hardening the log across AZs, replicating data to read replica's and standbys, handling dirty page writing, handling torn pages, etc.). The result is simpler algorithms and simpler code. This is the best way to make a system more reliable and more performant.

Everyone is trying to figure out how junior people will level up in a world where AI does the tasks we all used to do to learn. I don't think that 'simulated work' like the WSJ is reporting some companies are doing is going to be the winner. Curious what folks are trying though.

Love what we can do with Transformers, like safely approve more folks without much credit history.

@Affirm Impact on thin file users. These consumers have credit reports, but insufficient history for a FICO score. Here, the gain was 2.1x that of the next XGBoost candidate. A missing score doesn’t mean the underlying accounts and payment histories have nothing useful to tell us.