|
|
|
|
|
|
by a2128
1482 days ago
|
|
|
According to the model author, this is less about GPT-4chan being more truthful, and more about TruthfulQA not being a good benchmark. Possibly this result is due to the fact that the benchmark treats uninformative or irrelevant answers such as "No comment" or "It's raining outside" as being truthful. |
|
|