How does data quality affect my AI?

Your Personal AI relies on the accuracy and quality of the data you decide to upload. The expression “garbage in, garbage out” applies to this situation, as erroneous or inaccurate data has the potential to generate low quality responses.

Every piece of information you upload is used to train your SLM—be it good or bad data, everything is stacked in the same way. The training process is nearly instant, with the AI being ready to answer questions on uploaded topics in just a few seconds after. Then, when you ask a question, the first step of the request is sent to your SLM for processing: it’s at this point that it generates an answer based on your data. The rest of the message is padded with LLM text to improve fluency.

Deleting bad memories

If your Personal AI trains on inaccurate data, you can fix the issue by deleting the affected memories:

• Find and delete the documents, media or files in the platform that contain the affected memory. You can use the Training Studio to browse through the stacked data.
• Search your memory stack for the affected memory. Start by searching for the keywords—for example “we had an increase of 110% in new unqualified leads”—and browse through the results. Once you find the bad memory, you can delete it.

As you remove bad memories, you’ll notice your AI no longer relies on them to generate answers. Learn more on how to search for and delete bad memories.

Reinforcing good memories

Just like in LLMs—more common facts usually win over rare ones—you can push your Personal AI to respond consistently on a set of topics by uploading more information on it. The more memory blocks that reference the same fact in the same way, the higher the personal score will be, increasing accuracy and trust.

‍