Maybe an AI should automatically read the article and generate an image based on it. At least that would give more information than a random (albeit beautiful) color thing.
Not sure whether this is a good idea, but it seems like an interesting experiment. Perhaps you could try this with some selection of high-karma (so the readers are more likely to remember them) recent articles that did not contain a picture, and post the results?
If the prompt “draw something that is associated with the text of the article” is not sufficient, maybe it could include some nudges, such as “check whether the article contains math? is about AI? is fiction?” and then maybe make some specific recommendations (e.g. use a specific artistic style for fiction).
Yep, we considered it, but were a bit hesitant to editorialize too much on behalf of the author (so went with very unopinionated images vaguely in the LW house style). We might try adding default images!
Maybe an AI should automatically read the article and generate an image based on it. At least that would give more information than a random (albeit beautiful) color thing.
Not sure whether this is a good idea, but it seems like an interesting experiment. Perhaps you could try this with some selection of high-karma (so the readers are more likely to remember them) recent articles that did not contain a picture, and post the results?
If the prompt “draw something that is associated with the text of the article” is not sufficient, maybe it could include some nudges, such as “check whether the article contains math? is about AI? is fiction?” and then maybe make some specific recommendations (e.g. use a specific artistic style for fiction).
Yep, we considered it, but were a bit hesitant to editorialize too much on behalf of the author (so went with very unopinionated images vaguely in the LW house style). We might try adding default images!