Search papers, labs, and topics across Lattice.
2
0
3
Switching an LLM from an API to a consumer chat interface can degrade performance more than downgrading an entire model generation鈥攁nd standard API hyperparameter controls cannot reliably bridge the gap.
Current reliance metrics miss the crucial shift in cognitive effort distribution between users and AI, but this new "offloading score" finally captures how much work is *actually* being delegated to the tool.
