LLMs tell bad jokes because they avoid surprises
11 days ago
- #AI
- #LLM
- #Creativity
- LLMs struggle with comedy, art, journalism, research, and science because they are designed to minimize surprise.
- Jokes require a balance of surprise and inevitability in hindsight, which LLMs cannot achieve due to their predictive nature.
- Good stories and journalism also rely on surprising yet inevitable elements, making them challenging for LLMs to generate effectively.
- LLMs perform better in areas like software development where minimizing surprises is beneficial.
- Future AI advancements may require hybrid architectures that incorporate curiosity to seek and validate surprising truths.