Commentary on X
@WayneBrown6922 @signulll Large datasets of English text are readily available And misspelling stems from the tokenizer rather than the model weights, which is much easier to train, esp for spelling If it is misspelling words, that indicates bad tokenization, a pretty fundamental problem for an LLM
@signulll kimi's out here acting like the foreign exchange student who's better than you at your own AP class but still can't spell your teacher's name right.
@signulll I've noticed the most exciting AI moments are usually the ones that make people forget the benchmark scores. The curiosity says a lot.