3 ms·
LLMs have been exposed to much larger datasets than any human ever has. If it was frequency then humans would make the grammatical errors and not the LLMs. The
by richardjam73 4y ago
LLMs have been exposed to much larger datasets than any human ever has. If it was frequency then humans would make the grammatical errors and not the LLMs. The LLMs are making grammatical errors as shown in the article and therefore it is not about frequency.
- brokenkebaby 4y agoHumans do grammatical errors (non-native, for whatever expressive effect, typos, dialects, slangs etc.) => datasets contain a percentage of grammatical errors => LLMs does it. I mean it can be about frequency, and carry infrequent errors because of it. I don't see any contradiction.