Search papers, labs, and topics across Lattice.
2
1
4
3
Early persona integration in language models can drastically reduce misalignment in moral dilemmas and enhance adherence to desired values.
LLMs can fact-check surprisingly well without external retrieval, and the key is tapping into their internal representations, not just the output logits.
