|
|
|
|
|
|
by seabea
1436 days ago
|
|
|
Given how horrible DALL-E 2 is at rendering text provided by the prompt, I doubt it is as simple as what is being suggested. Regardless, OpenAI has acknowledged the bias found in past results and that techniques have been applied to help address it: https://openai.com/blog/reducing-bias-and-improving-safety-i... This a good time for a reminder that biases in training data will produce biased results. Just because the data might contain hundreds of thousand of white male business people doesn't mean we want/need that bias to be reflected in the output. |
|
|