Alignment has a Fantasia problem
A paper from MIT names the failure mode where an LLM agent commits to an underspecified user intent and then faithfully executes the wrong thing. We turned the prescription into a routing skill.
Read the post →
Practice · Blog
Notes from the practice.
What the research papers actually say, what we shipped to test them, and what we'd build with another week.
More posts landing soon. Subscribe via the contact form to be notified.