4 ms·
More "quality" data equals more better. The argument here is the LLM generated text is now going to enter the corpus, muddying the waters and reducing the qual
by gibolt 3y ago
More "quality" data equals more better.
The argument here is the LLM generated text is now going to enter the corpus, muddying the waters and reducing the quality.
- chaxor 3y agoPerhaps Meta will release their dataset for Galactica sometime, and we will have incredibly good training data quality.
- dontupvoteme 3y agoCurrent LLM outputs already kind of reek of reddit-style responses, at least to me. Is a hivemind really any different than an AI making its own data?