As with all AI systems (and humans!), RunLLM will occasionally miss details or get things wrong — either because a concept wasn’t in the documentation or because we misunderstood the data we were given. While these mistakes are tough to avoid altogether, RunLLM makes sure that it never makes the same mistake twice with instant learning.

Instant learning refers to RunLLM’s ability to learn from supplemental data that you and your team provide within a few seconds.

When to trigger instant learning

Whenever you see a situation where RunLLM doesn’t provide an answer — like in the example above — or answer that’s missing detail and nuance, you can teach RunLLM with instant learning. The information that you provide will be combined with what RunLLM’s already learned from the rest of the data sources you’ve provided in order to increase the quality of future answers.

It’s worth noting that this information will be prioritized more highly in future answers than the other data sources you’ve given us. Since we weren’t able to find an answer in the data sources you’ve given us, we’ll use this data to answer future questions.

How to trigger instant learning

<aside> 💡 Instant learning is currently only supported on the RunLLM Slackbot. If you’d like ot use instant learning elsewhere, please reach out!

</aside>

Triggering instant learning is easy. When you see an answer that you’d like to improve, you simply need to:

1. Type out the correct answer and send it as a follow-up in the Slack thread with the original question.
2. Add the 💡 emoji to your message.
3. Wait a few seconds for RunLLM to learn and try again!

When typing out the correct answer, we recommend providing a detailed explanation of how the answer could have improved and even providing links to relevant source documentation. These two principles will help ensure that RunLLM learns as comprehensively as possible for future answers.

Who can trigger instant learning

Given that teaching the assistant directly is susceptible to malicious attacks where incorrect or irrelevant information is fed into the assistant, RunLLM restricts who can trigger instant learning.

By default, instant learning can only be triggered by Slack users who have an email account with your company’s domain. If anyone else uses a lightbulb emoji on a message, that action will simply be ignored. This ensures that only trusted company staff can teach the assistant and avoids a scenario where an attacker could teach the assistant incorrect information.

We’re working on other modes of authentication, and if you have a configuration you’d like to see, please reach out!