Search papers, labs, and topics across Lattice.
1
0
3
LLMs can rediscover known algorithms, but only after targeted unlearning and with the help of a generative verifier to avoid "thought collapse," revealing both the innovative potential and limitations of these models.
