3 ms·
the LLM were already trained on data with toxicity, This is why they have the weightings to know how to respond. They aren't really teaching the model anything
by s1artibartfast 3y ago
the LLM were already trained on data with toxicity, This is why they have the weightings to know how to respond.
They aren't really teaching the model anything it hasnt seen before.
If you want toxic output, you can get it.