Search papers, labs, and topics across Lattice.
Harvard University, Kempner Institute at Harvard University
3
0
6
HB doesn't push the compute efficiency frontier beyond SGD, but it does extend the batch-size window for reduced serial runtime significantly.
Language models can bootstrap their reasoning abilities without human labels by learning from each other's aggregated answers, achieving significant gains in mathematical reasoning.
Language model capabilities are surprisingly stable over time for most tasks, except for math reasoning, which continues to advance, offering a way to reliably translate compute budgets into performance expectations.
