What can and can’t you do with Affino’s Enterprise Expert AI Service

AI will impact on virtually all aspects of work and entertainment over the years ahead. In practice this will be a multi-layered approach with complex systems having dozens or even thousands of different expert AI services coming in to play to perform specific functions. General AI’s are often to broad in their remit, so having focused AI services is currently the best way to have the most effective outcomes.

 

Things Change Fast

 

It is worth checking in on this insight piece regularly because in the world associated with AI (Artificial Intelligence) things change very fast.

 

Every few months there are considerable leaps in AI capabilities which means that the services which can be built on them also improve rapidly.

 

Since launching Affino’s earliest Expert AI services we have seen the rate of hallucinations decrease massively. Things have gotten much faster. The presentation is much better. The chat interface is more intuitive. And the expert AI services themselves are greatly improved since the tools for training them and managing the feedback are so much better. The analytics are clean and simple, so the overall learning curve is kept pretty low for getting started and delivering fully fledged services.

 

We always tell people that Affino is evolving quickly, with over 1,200 feature updates each year. By fully integrating AI services, we open up a whole new world of capabilities that will take this a great deal further and faster.

 

What can you do with Affino’s expert AI Service

 

Affino’s approach is to focus on a chat style interface, and leverage existing expertise to deliver great new services.

 

We do this by building effective tools for you to import your existing text based content to Affino, to manage it within Affino, and to use it to train up and deliver expert services.

 

In practice you can’t simply import your content and have everything work smoothly from day one. You need to work on and evolve your expert AI services to add in a level of safeguards, correct misinformation or out of date information. Train up the AI to help it give the best answers and give them in the best sequences.

 

What can’t you do with Affino’s Expert AI Service

 

Our current focus is on text only, but this will change to add in images, audio, video and other media formats over time.

 

It still feels like early days for many of the image and video API’s, and we are waiting for them to mature a bit, as well as give us space to focus on the text side as that is where most of the world’s knowledge still resides.

 

Affino’s AI is also purely delivered through a chat-style interface, and in the future this will likely evolve to provide many more interfaces but chat is a great place to start.

 

What can Expert AI Services be great at

 

There are quite a few elements which make up a great Expert AI Service, just one of which is the LLM (AI) behind it.

 

You can get into the details by looking at the overview of Affino’s Expert AI Services here.

 

At its core you can get all the content easily into Affino, you can see exactly why your AI service provides each answer, and you can tune that very easily by updating the content or creating quick FAQ’s to provide vital additional information and ‘bridging’ information for creating the best service.

 

Affino’s Expert AI Service helps you to deliver and monetise expertise and entertainment to your audience.

 

For professional audiences this means providing expert information and insight in a continuously improving way, where every answer provided can be analysed and improved on over time.

 

For consumer audiences this means providing the best lookup tools, creative story telling and learning experiences, and many more.

 

Some Caveats on how Affino’s Expert AI Services Work

 

There are many things that the Large Language Models (LLM’s) are not net good at, but which will come in time. Here’s a useful list, which we’ll work to keep up today as the AI models evolve.

 

This goes for pretty much all LLM’s currently, with some being better at certain things and some being worse.

 

Currently the best overall service is OpenAI’s ChatGPT so we primarily use their API at this time for the AI specific tasks.

 

Things LLM’s are not good at

 

We expect these things all to improve greatly over the coming year but these are the biggest things to keep in mind when rolling out expert services on the current AI limitations.

 

Math(s) AI’s can be very random when handling mathematical problems, and this varies a great deal between LLM’s but the bottom line is you can’t 100% trust this yet.

 

Understanding the sequence of time, e.g. what’s news. In Affino we timestamp all the content we feed into the AI database and provide an option just to search within the new content. Asking an AI who is the current CEO of a company is very unreliable unless there is a specific well referenced bit of text saying exactly who it is.

 

AI’s do still hallucinate a little, but often not for the reasons you might think. For example if A and B are possible, it might think that doing A is possible, e.g. you can add content and export from a system, so the AI might think you can export all content (when in practice only exporting specific types of content is possible). You will need to train it up to know the specifics better.

 

If you have 10,000 reviews and ask the AI how many of each type of widget did you review, without feeding every review into the LLM each time to be processed, or providing the specific total to the AI, the AI will simply not be able to know this. Commercially it would be very expensive to get the AI to interpret all the reviews each time a question is asked, even if it can handle it (in the context window).

 

We would expect all of these to change in time. We notice that with each iteration of AI it is not simply the case that things get better, often new issues arise and new constraints are put in place by the AI model which then need to be addressed. We’ll keep this list updated or create updated lists as things evolve.