Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
17
ES not only outperforms GRPO in reasoning tasks but also reveals that substantial parameter updates don't equate to widespread functional changes in LLMs.
