tech ๐ง Big data pools wash away AI's theft claims
A new MIT paper suggests visual similarity in AI output isn't the same as causal theft from training data. Researchers found that as datasets grow, the influence of any single artist's work drops towards zero using 'ablatable ensembles'. This provides a scientific shield for diffusion models like Midjourney and OpenAI. The findings apply to image generation, not text, leaving LLMs open to different copyright debates. The real test remains how these models are used in actual court cases. ๐ผ๏ธ