Your network blocks the Lichess assets!

lichess.org
Donate

This is not a Kitten

You could blame the big data mess that trained the LLM to picture the 2 ladies in the back not being players at the table. It might represent more movies images (but are they even trained with internet images on to have word to image training, is that big data too, of the same uncharacterized level that is called big data?).

Like all the hallucinations, those creative images, represent the training set. It is just that there might be streaks in there or lack of diversity (not the buzz word, but the buzz word is also about diversity), should I say high cardinality types, of images..

if it is movies that are used, how often are chessboard seen, mostly about protagonists... how often are the protagonist female in movies that play chess.. in the big data mess I mean. in proportions..

So whether it was inteded or not, I think using the same "elevator music" target of learning machine, as a non human creative helper, is likely to keep repeating a lot of stereotypes... or just limited numbher of types..

It might need the same amount of work as creating one from scratch to substract things. And we are fallible, give the time we might put on this, it might be better to show that (from scratch), than have some mess machine express our individual need to share our thoughts, and then putting one self in every possible angle of the audience that might be offended by the rich amount of recognizable things but not our deliberate choice to be there.

I think you have found a solution. No more distractions from the big mess. Better have crooked visuals.
I might put back a blog of mine that I "hand" drew using excalidraw.

But on the other hand.. If using LLM based technology (hooked to visual diffusion stuff, and ? image training set) why not have an appendix or outlink to the prompt conversation or the single prompt that did the job. Then the offending or offended matter would be attributed to the responsible causative agent.

Lots of work from an inidividual. More distraction from the content.

so back to your solution. or crooked pictures.. Think of lichess horsey.. There is some value there, in my eyes (which might also be crooked, inside, the mind's eye(s)).

You could blame the big data mess that trained the LLM to picture the 2 ladies in the back not being players at the table. It might represent more movies images (but are they even trained with internet images on to have word to image training, is that big data too, of the same uncharacterized level that is called big data?). Like all the hallucinations, those creative images, represent the training set. It is just that there might be streaks in there or lack of diversity (not the buzz word, but the buzz word is also about diversity), should I say high cardinality types, of images.. if it is movies that are used, how often are chessboard seen, mostly about protagonists... how often are the protagonist female in movies that play chess.. in the big data mess I mean. in proportions.. So whether it was inteded or not, I think using the same "elevator music" target of learning machine, as a non human creative helper, is likely to keep repeating a lot of stereotypes... or just limited numbher of types.. It might need the same amount of work as creating one from scratch to substract things. And we are fallible, give the time we might put on this, it might be better to show that (from scratch), than have some mess machine express our individual need to share our thoughts, and then putting one self in every possible angle of the audience that might be offended by the rich amount of recognizable things but not our deliberate choice to be there. I think you have found a solution. No more distractions from the big mess. Better have crooked visuals. I might put back a blog of mine that I "hand" drew using excalidraw. But on the other hand.. If using LLM based technology (hooked to visual diffusion stuff, and ? image training set) why not have an appendix or outlink to the prompt conversation or the single prompt that did the job. Then the offending or offended matter would be attributed to the responsible causative agent. Lots of work from an inidividual. More distraction from the content. so back to your solution. or crooked pictures.. Think of lichess horsey.. There is some value there, in my eyes (which might also be crooked, inside, the mind's eye(s)).