-
#hacking!
-
ln2403 embraces the new bible
-
look at them go
-
productive to an unreasonable degree
-
juan + team cap
-
Architecture map
Inspiration
Lack of AI transparency and accountability for the general public in decision making processes is on many of our peers' minds for obvious reasons. We were inspired by forward-thinking innovators in our communities with minds open enough to see the opportunities that come with change and to do the work to be agents of it.
What it does
LANTERN is a prototype AI model evaluator that
How we built it
Challenges we ran into
A lot of the dirty work was keeping the repo straight while multiple sessions were moving at once. We had old worktrees, unfinished branches, staged changes, and handoff docs that no longer matched what had landed. We had to check what was actually merged, keep active work separate, and make sure cleanup didn’t wipe out someone else’s changes. Getting the demo ready also meant separating the real pilot from the synthetic frontend fixture so we showed the judges the right data
We used issues as a task board for ourselves, and for our agents. This helped us overcome this challenge by enforcing a strict git-control paradigm and enforcing contracts with our agents. For the most part we were able to avoid merge conflicts through this method.
Accomplishments that we're proud of
We got the full pipeline running, retained all 20 real responses, and built a demo judges can inspect even when the result is that nothing changed.
We got inference running on Alpha and connected its outputs to LANTERN’s recording and comparison workflow.
What we learned
What's next for LANTERN
LANTERN's architecture is designed to be flexible. Our original concept was journalistic in spirit. We envisioned the architecture at the heart of an engine that could bring light to AI decision makers in markets and other human systems, tracking patterns of behavior to determine (broadly speaking) whose interests were served by its actions. We quickly realized the core functionality could be applied diversely to big-picture issues regarding artificial intelligence and chose to reinforce its foundations. In the future we see LANTERN providing the analytical underpinnings of a bunch of stuff.. all the stuff that benefits from following the bread crumbs of high-powered automated decision making. model for the long-term and far-reaching consequences of
