Starvard
Prompt Engineering Guide · Research
LLM In-Context Recall is Prompt Dependent
This new paper by Machlab and Battle (2024) analyzes the in-context recall performance of different LLMs using several needle-in-a-haystack tests.
It shows that various LLMs recall facts at different lengths and placement depths. It finds that a model's recall performance is significantly affected by small changes in the prompt.
Source: Machlab and Battle (2024)
In addition, the interplay between prompt content and training data can degrade the response quality.
The recall ability of a model can be improved with increasing size, enhancing the attention mechanism, trying different training strategies, and applying fine-tuning.
Important practical tip from the paper: "Continued evaluation will further inform the selection of LLMs for individual use cases, maximizing their impact and efficiency in real-world applications as the technology continues to evolve."
The takeaways from this paper are the importance of careful prompt design, establishing a continuous evaluation protocol, and testing different model enhancement strategies to improve recall and utility.