2 ms·
Can the AI tell the difference between quality and shitty memes?
Right away I'd say — not really, as ChatGPT knows 25 jokes and you heard them all before, but I thought I could make it work another way. People have tried to use ChatGPT to generate memes (supermeme.ai), but I decided to try make AI differentiate between quality and shitty memes.
The AI will just aggregate memes and send them to users, in turn users share their reactions and help the AI learn about their preferences.
The main challenge was to build a solid RecSys. I initially tried to use Amazon Personalize but it didn’t work as expected — I got similar results just using a random. I dropped it and built my own RecSys based on matrix factorization to make a v1 of the algorithm. I packed into a telegram bot & called it Lolla (telegram.me/Lolla if you fancy see the result)
Do you think something like this has potential or trying to teach AI to detect quality memes sounds more like a load of bs?
- seanthemon 3y agoThis is just a classification problem. You can use a standard neural nerwork for this, the input parameters are the pixels of the meme, the expected output is GOOD or BAD you can check huggingface to see if any models could be fine tuned to handle this or you can use tensorflow[1] You will need quite a bit of good data for this to work, GPT 4 can work with images and you could fine-tune it but i'm not sure how to do that for images. As to usefulness - to me it's not really, but as a model it could be something fun someone can use or find use of [1] https://www.tensorflow.org/tutorials/images/classification https://www.tensorflow.org/tutorials/images/classification