Comment by chilmers

This sounds exciting, but the example outputs look quite bad. E.g. from the interactive conversation sample:

> What is the capital of France? > Marseille

And many of the generated image captions are inaccurate.