Hello again and welcome back to AI Corner, your one stop shop for all things Artificial and (hopefully) Intelligent. Today I’m going to go over some common pitfalls when it comes to using the AI Assistant and offer some tips for getting the most out of it.
Plus, I’ll also discuss some of our recent updates to our AI Assistant that expand what you can do with it. Let's get into it...
The SyncHub AI Assistant is powerful. The AI models powering it are smart and effective but, ultimately, they produce their answers by picking the most likely next word one after another. To get the best results you want to stack the deck in favour of your intention, so the name of the game is reducing ambiguity. The more uncertainty there is in a request the more likely it is that a model will misinterpret. There are a few ways you can improve this:
Be precise
The simplest way is to make sure you are as precise as possible in your requests. Because of the very human-sounding way that AI writes, it’s easy to fall into the trap of thinking you are talking to something equivalent to a human BI analyst. There are some critical ways that this is not true, one of the biggest is that (unlike a human analyst) the AI will very rarely push back for clarification. Instead, they will almost always simply proceed with the most likely (in their view) interpretation.
Anything you don’t tell the AI Assistant is something it has to decide for itself
For example, if you ask something like “Who is my most valuable customer”, there is actually a great deal of nuance in what “most valuable” might mean, so narrow it down or explain what “most valuable” actually means to you. Is it just the sum of all invoices in a certain time or are there other factors? Does longevity matter? Does reliability of payment? Anything you don’t tell the AI Assistant is something it has to decide for itself and every decision increases the chance of a mistake.
AI needs broader context about your business
The AI Assistant knows the structure of your data and how items link together, it knows details about what different properties and fields mean and how they can be used and it knows how to put these pieces together to retrieve the information you ask from it.
However, without some input from a user it doesn’t know much about you. It doesn't know what your business is or what your goals are, it doesn’t know how you like things arranged and it doesn’t know all the little details about the way you use your cloud services that are different from the standard user.
That’s where Personalization comes in, allowing our AI Assistant to remember important details in every conversation. There are two ways to use this:
- The first is by directly writing them in the personalization window, with SyncHub. To do this open up the Chat window, then follow the prompts to Personalize your AI. A new window will open up that lets you enter any details you like either about your business in general, or we event let you link your notes to a specific table or column.
- The second option is to ask the AI Assistant to remember something, directly from within your chat. This is simpler but it will rely on the AI’s judgement to generate the actual memory and put it into the appropriate context (see above), so if you want precise control of the memory its always better to write it yourself.
Pick the right model for the right job
It’s important to pick the right AI model for the right task, and SyncHub supports integrations with all the top-tier providers. Across these, there are many models available that are more or less sophisticated and more or less tailored to certain jobs. For example, Claude models are exceptional at code generation including writing SQL, while Gemini Flash or Pro can be better at interpreting human intention.
Under the hood the Assistant breaks down the questions you ask it into discrete and isolated chunks each of which is handled with its own context and instructions, and now, thanks to a recent update, it’s own model.