Search papers, labs, and topics across Lattice.
1
0
Training spikes in LLMs are not an emergent property of massive scale, but can be replicated by simple quadratic models with power-law inputs.
