By Marc Stephenson |

October 4, 2024

AI is still all talk and no action

Everywhere we look, we seem to be bombarded with articles about Artificial Intelligence, AI architecture diagrams, numerous AI use cases and how it will revolutionise our lives. However, have any of us actually experienced the real value of Gen AI in practice?

In this series of blog posts, we’ve been taking a deep dive into the world of AI. For our latest blog, Marc Stephenson, Director here at Metataxis, shares his thoughts about the current AI situation:

The value of architecture diagrams

I love an architecture diagram. I often create them, to good effect, as part of my consulting work.

Architecture diagrams clarify your thoughts and intentions as to what needs to happen, and they can be effective at explaining solutions to clients.

However, if I see another Generative AI or Large Language Model (LLM) architecture diagram – I will scream!

I get it. I understand how Gen AI can be architected for RAG (there is something about the acronym for Retrieval-Augmented Generation that bugs me – but we’ll save that for another day!). I also understand what “grounding” is and how it fits into LLM solutions. I even understand how the innards of an LLM actually work.

Large Language Model diagram

Is Gen AI really at work?

What I am yet to see (not in diagram form!) is an actual, real-world, out in the wild, working Gen AI solution. Something that actually delivers specific value and does not simply show me how easy it is to summarise a meeting or lengthy report.

I’ve heard that there are some real solutions out there, but I’m yet to see them. Slick demos (however useful looking) from vendors that speculate on real use cases do not count.

Has anyone seen an actual, real-world, out in the wild, working Gen AI solution?

Don’t get me wrong. I’m not an AI sceptic. I use Gen AI every day, mostly the paid for subscription to ChatGTP. For me, it has replaced most of my search engine use. And I still find it alarmingly good at general queries, especially when you can validate the answers easily – this is why it’s excellent at generating code.

How will AI perform with actual customer data?

However, I’d like to see a working implementation of RAG that delivers value to potential clients, using real, messy, client data. Microsoft Copilot, in theory, should be able to do that with the information stored in my Microsoft 365 tenant. Copilot should help me be quicker and more efficient at finding and leveraging information in M365. But I remain, as of yet, unconvinced.

Maybe I just need to be a bit more patient and all will be fine once Copilot is available on my tenant, and I can see how Gen AI will really, really benefit me.

Have you seen Gen AI in action, yet?