Neil Ernst

In an era where AI can generate plausible looking marathon finishing photos, plausible code for code review, and plausible 10 page conference papers, the challenge will be for us humans to prove the veracity and value of our work.

That we really did *run and finish* the marathon, we actually *know* what is in the code, or really *conducted* the research in the paper we submitted. My level of trust in these artifacts - by themselves - is at rock bottom.

My kids were in a hackathon over the weekend building a game with Godot. Their biggest challenge: merge conflicts with Github. They took solace in the fact even senior undergrads (or professionals for that matter!) struggle with Git and Github.

"Train Dreams" is a fantastic meditative film, with understated acting and rich cinematography. I'm biased as someone who grew up in this landscape, but I think it appeals broadly. 10/10

New blog post on our TACL paper: https://neilernst.net/posts/topic_models_tacl.html

We did some human experiments with how people interpret topic models. We found traditional statistical metrics don't capture the variability human judgment brings. People anchor on key words in the topic list, for example.

I also think there are important implications for how GenAI tools similarly interpret distributions of words.

Can't believe it needs to be asked, but US folks please consider this letter campaign: https://agu.quorum.us/campaign/157636/ It is about supporting the vital work of NCAR and NOAA and others on climate science.

A lot of my day seems to now consist of pressing `1` or `2` after parsing Claude's suggestions.

Reminding me of a Simpson's gif:

Wonderful talk this afternoon from Rek and Devine, the 100 Rabbits, and force behind (among many other things) Uxn - https://100r.co/site/uxn.html Uplifitng and thought provoking and just a good way to end the week.

Every time I hear about OpenClaw I think Malazan Book of The Fallen and the Claw assassin guild.

Would be way cooler. Who knows, maybe OpenClaw will also start knocking people off, given its total lack of guardrails.