Results of AI field testing

[Dawn Wolthuis]

For those who have been following the AI advances out there and have noticed how many software companies now tout their AI, I have a tidbit to pass along regarding the acceptance of AI when field testing it in several organizations.

We put OpenAI into Entrinsik's Informer BI tool, and it seemed amazing to us, so we were excited to field test it. Well, turns out it was amazing, but some of the key use cases had gaps -- too many. With the results back in the hands of our development team, we came up with a new feature set that hit the nail on the head for these use cases: AI Assistants.

These AI Assistants are scoped chatbots, of a sort, that are created handily by a site (or by us, if preferred) where you identify the datasets they will use and the unstructured data in pdfs that you want used, along with any other rules you have for the AI assistant. You can now build these atop your reports/BI in Informer, with proper security. We have done a lot to minimize and mitigate AI hallucinations too, as those are not welcome with data questions.

If you did not see the webinar, the recording is here  --https://entrinsik.com/informer/recorded-webinar-unlocking-the-future-transform-your-work-with-informer-ai/

Apparently Amazon liked our idea and now also has AI assistants [OK, there is a chance they did not get that idea from us]  https://aws.amazon.com/q/  Those require building from scratch all of the reporting/BI infrastructure tightly coupled with AWS if you want to get at your data for the AI to answer questions. Informer is a faster zero to hero lift for AI, I suspect, with role-based column & row security already set up with the BI features. Admittedly I have made an Informer AI Assistant (yes, even I!) and not one with AWS.

[AD] If interested in seeing AI with your data, we do free trials. You can either put in a request on this page https://entrinsik.com/informer/ or just set up a time for an initial chat with me here https://calendly.com/dwolthuis.

I love to hear what organizations are hoping AI can do for them as well as what their fears are regarding use of AI. You need not have an interest in Informer to chat with me about AI -- just pick a calendly time so I can hear your thoughts on this topic.  Thanks.  --Dawn

https://entrinsik.com/informer/ 

[Tony Gravagno]

Greg Howe (also a former MVer, now shifted to AI as I have for the most part) posts frequently in LinkedIn on this topic and offers related consultation and services. I encourage everyone to check out his posts there, learn from a colleague, and translate his tips into your applications for fun and profit.

From my point of view, here is the rough evolution that I think people go through with this:

- ChatGPT : The Hello World of chatbot applications and OpenAI's gateway drug to the realm.

- CustomGPTs : Entry-Level opportunity for people to create a custom knowledge bot or creative agent with optional code interface. (My wife and I create these for fun.)

- Completion API : First Hello World stop for all developers to code an exchange with a LLM.

- Assistant API : First step beyond "wow, that was cool". Essentially Custom GPTs via API and without the UI.

- Embeddings : Next step toward "how do I use this with my own data" ... On-ramp for databases and knowledgebases.

- RAG : Serious developers' paradigm for really using the tools interactively with data and logic.

- Agents : Pro stuff - making it look intelligent without it actual AGI. Workflow, decisions, async processing : Borg Collective.

- Multi-modality - Enhancing UIs with audio I/O, image I/O, now video I/O, and more...

"AI" is still just a Marketing buzzphrase. All it is now is text-processing, predicting which word should come next in a sentence. There's no "intellegence" at the application level. The current technology only looks smart with RAG and Agent architectures. In a year or two it will be different with AGI, and the average person might be a little closer to separating one concept from another. Where it is now, it's still relatively easy to hop on the train, and if your clients will pay for that, there should be no objections to providing it.

Most of the products that claim AI these days are based on some variation of the Completion API. As Dawn said, the serious players soon realize that there is much more to this, and they move forward. I saw the LLM interface in Informer some months ago at Dawn's invitation, and it was truly impressive. I encourage everyone to take a look - it will only get better. If you want to offer your clients the letters "A.I." and you don't want to have to write code, definitely look into what Entrinsik is doing with Informer and "smart" reporting.

In my past life here I would be talking about AI with MV. Today, I note that there's near zero mention of the topic anywhere in the MV industry, which is exactly why I had to shift away. That continues to be painful and I wish it were different. On the bright side, as with all other tech that we have encountered over these decades, I just want to convey that this is yet another technology that absolutely can work with any MV-based application. Just do your market research to identify demand and cost-justification, then use the same techniques that I've talked about over the years.

For all of us who have trouble figuring out where to hop on the train and what to expect, as always I'll be direct. This one isn't easy by a long shot - but it's still do-able. The APIs and capabilities of all of this stuff are rapidly changing. It's tough to keep up because New becomes Old within weeks. Best Practices and Design Patterns are evolving very quickly - as soon as you catch on to "this is how it's done" it changes, and there are suddenly better ways to do things and more platforms that claim to be better suited. For example, Fine-Tuning was the defacto practice for data-enabling a LLM, until the disadvantages were quickly and commonly recognized. Embeddings became popular, and now as people shift away from the LLM (and GPT) as being the main focus of all of their attention, multi-LLM environments with are common to support even small applications. RAG just means "data integration" and it's becomming a sub-industry in its own right. Most people aren't into agents yet. It's very complex and too new - there's a high investment time/cost, and because it changes so quickly it's difficult to get a stable and durable solution into production quickly that will provide ROI for the next cycle.

If you feel a sense of encouragement and discouragement, keep up with Greg. He can help to guide management and technical staff.

HTH

T