What prompting an LLM can do
- Can a fresh college graduate follow the instructions in the prompt to complete the task?
Knowledge cutoffs
- An LLM's knowledge of the world is frozen at the time of its training(2022年1月)
Making things up: Hallucinations
- so understanding its limitations is important if you are using this for documents of real consquence
The input (and output) length is limited
- Many LLMs can accept a prompt of up to only a few thousand words.
- The total amount of context you can give it is limited.
- Some LLMs have longer context limits - up to 100,000 words
- An LLM's context length is the limit on the total input+output size
Generative AI does not work well with structured (tabular) data
- use supervised learning to structured data
Generative AI works best with unstructured data
- ( text images, audio, video)
Bias and Toxicity
- An LLM can reflect the biases that exist in the text it learned from.
- Some LLMs can output toxic or other harmful speech, but most models have gotten much safer over time