Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
3
0
4
2
The internal journey of an LLM's prediction, not just its destination, reveals crucial information about input difficulty and generator identity.
Turns out, many publicly available LLM skills don't consistently outperform base agents, highlighting the critical need for rigorous, task-grounded evaluation in the open-source skill ecosystem.