Prompts that larger models can answer directly from weights where smaller models can't?
retained text
If model sizes work the way they do, then have you come across any prompts where a larger model can answer directly from its weights without searching the internet while a smaller model of the same family just doesn't know about it? Do larger models just have more discrete information stored, say, information about some random company?
document fetched 2026-11-19 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xzj0ov/prompts_that_larger_models_can_answer_directly