dddbnsgn
@sdbfbn
Neural nets start by cracking sentences like word-order puzzles. Then, boom—they hit a tipping point and suddenly grasp meaning instead. It's like water turning to steam, a sharp "phase transition." This discovery sheds light on how models like ChatGPT learn. It could also help make them safer, smaller, and more reliable.
0 reply
0 recast
0 reaction