To collaborate effectively with your Olympia assistants, it's crucial to grasp the concept of their "context," or attention span, during conversations.
Context encompasses both the immediate conversation and your broader interaction history with the assistant. This is what enables Olympia's AI to understand your intent and respond in a meaningful way. It takes into account not just the current conversation but also relevant details from past interactions and even your initial onboarding interview.
In the realm of AI, a token is a sequence of characters that forms a unit of meaning. The assistant's context—or its attention span—is measured by the number of tokens that the underlying language model can consider while generating a response. Olympia's AI models can handle up to approximately 5,000 tokens while maintaining high levels of focus and intelligence. Depending on the length of your prompts and the assistant's responses, this could equate to just a few or dozens of messages.
When you give the AI a long prompt, it will take its cues on how to respond to you from whatever came last in the prompt, not what was at the start. This is a super important understanding to have when you’re entering long prompts.
For example, let’s say that you have some terms and conditions language for a contract, and you want our AI to simplify the writing. The right way to accomplish your goal is to enter the terms and conditions first, then hit enter a couple of times and write your instructions. The AI will take the terms and conditions you want rewritten into its context, and then see your instructions at the end. If you put the instructions at the start, before the terms and conditions, and the terms and conditions seem like they might be incomplete, the AI is as likely to just continue “finishing” the existing terms and conditions writing you provided instead of rewriting the whole thing.
<aside> ⚠️ When entering a long prompt, make sure to put your instructions at the end of the prompt, not the start.
</aside>
One effective technique for enhancing your assistant's performance is to start a new conversation each time you switch topics. This allows the assistant to establish a fresh context, which is essential for better understanding your intent and providing accurate responses. Just like in human conversations, abruptly changing topics can lead to confusion; the same holds true for AI.
<aside> 😅 In contrast to other AI chat tools like ChatGPT, your Olympia assistant retains recent observations in the context of new conversations, and is capable of searching its long-term memory for relevant information to the topic at hand. Starting new conversations should not be painful or involve retyping previously entered data!
</aside>
Starting a new conversation not only helps with context and understanding, but it also helps with organization and clarity. By keeping conversations focused on specific topics, it becomes easier to locate information later on if needed.
If a conversation exceeds the token limit, Olympia employs methods like recursive summarization to condense the existing context while retaining important information.
Messages that have been summarized will appear as "archived" and are grayed out.