Analysis
This fascinating article explores the behavior of a 'base model' before alignment, providing a unique glimpse into the raw potential of Generative AI. By experimenting with a local Large Language Model, the author discovers surprising outputs and highlights the impact of alignment training. It's an exciting exploration of what AI can do before it's been 'taught' to behave.
Key Takeaways
- •The article showcases the difference between a raw Large Language Model and its aligned counterpart.
- •Experiments with the base model reveal unexpected and creative outputs, such as anime reviews.
- •The ease of local experimentation with Open Source tools like Ollama is emphasized.
Reference / Citation
View Original"The Base model, for 'hello,' isn't a greeting. It's just a Japanese token string."