RESEARCH

Am old enough to remember when @GeoffreyHinton told me I was stupid for saying that LLMs regurgitate training data. He was wrong. LLM regurgitation is now one of the best-established findings in the field. Excerpt below from a new DeepMind paper; every single one of the

→ View original post on X — @garymarcus

"Iterative Finetuning is Mostly Idempotent" If a model has a behavioral tendency and you train the next generation on its outputs, does the tendency snowball? This paper actually finds that it's mostly no. So they seed models with traits like sycophancy and misalignment, then

→ View original post on X — @askalphaxiv

"Demystifying Manifold Constraints in LLM Pre-training" LLM pre-training depends on a stack of heuristic tricks, eg. normalization layers, weight decay, and optimizer tuning. This paper suggests that by constraining weight matrices to geometric manifolds like spectral or

→ View original post on X — @askalphaxiv