Best Practices
Learn about the best practices for getting the most from your Neuro+ subscription. Below is a high level overview, the following chapters will go further in depth.
Using Neuro+'s Large Language Models (LLMs)
When working with LLMs, it is essential to understand prompt engineering techniques to effectively communicate your intent and guide the model's output. Crafting clear and specific prompts is crucial for obtaining relevant and accurate responses. Additionally, be mindful of the model's context length, as LLMs have a limited capacity for retaining and processing information.
Prompt Engineering
Prompt engineering involves carefully crafting the input text to elicit desired responses from the LLM. Some tips include:
- Provide context and background information
- Use specific examples or analogies
- Break down complex tasks into smaller steps
- Experiment with different prompt formats (e.g., question-answer, fill-in-the-blank)
For a more in-depth explanation with examples, read more about prompt engineering here [link to prompt engineering docs page]
Tips and Tricks
- Leverage few-shot learning [link to prompt engineering] by providing the model with singular or multiple examples of desired outputs
- Iterate and refine prompts based on the model's responses
- Combine outputs from multiple prompts or models for improved results
- Experiment with different temperature settings to control output diversity
Model Differences
Different LLM providers (e.g., OpenAI, Anthropic, Google) offer models with varying capabilities, strengths, and weaknesses. Some models may excel at specific tasks or domains, while others may have better performance on open-ended prompts. It's essential to understand the unique characteristics of each model and choose the most suitable one for your use case.
Different providers will also have different strengths and weaknesses. As an example, OpenAI's GPT-4-Turbo has comparable coding ability to Anthropic's models, but Claude 3 Opus may be a superior model to use for writing more human sounding communications and purely text content.
You can read more about model differences here: Model Differences [link to article]. For more information on the individual models we provide, you may find them here: [link to Available Models]
Context Length
LLMs have a limited context window , meaning they can only process a certain amount of text at a time given inherent model and hardware constraints. Longer inputs may be truncated or ignored, potentially leading to inaccurate or incomplete outputs. To mitigate this, you can break prompts down into smaller chunks or leverage techniques like context windowing or hierarchical prompting.
When setting context length, a conservative rule of thumb for the number of equivalent words is context length divided by 5. So a model with context length 50000 can approximately process 10000 words at a time.
You can read more about context length here: [link to context length article](Context Length).