resources
notes
- tokenization
- Base vs. instruct model
- ‘Regurgitation’ vs. ‘hallucination’ - when an input is in the training data vs. not
- Few-shot prompting is effective, pattern recognition, in-context learning
- ‘Recollection’, neurons, memory
- Probing a model to find out what it doesn’t know
- Programatically adding new examples…?
society
cognitive offloading https://jzhao.xyz/thoughts/LLMs