You would have to be living under the proverbial rock to not have seen the hype around AI in recent years. Satya Nadella, Microsoft CEO, describes it as a tidal wave as big as the internet, whilst Google CEO, Sundar Pichai, warns that society will have to brace itself.

If you are in the market for hyperbole however, this isn’t the place for you. Hymans is running a series of pilots to learn how AI tools can be used safely, ethically, and effectively. To learn more about this, talk to professionals and discover innovative technology, a couple of our Technical Architects recently headed to Microsoft’s AI Build Day at the ExCel in London.

Format

The day was nicely broken up into a keynote session from Scott Hanselman, a series of sessions of several topics such as Using AI in the Cloud, Responsible AI, and using tools such as Copilot.

The afternoon was given over to hands-on labs, which were a great opportunity to try out some of the things we had learned, with experts in the room to ask follow on questions. I found these sessions particularly helpful.

Key take homes

For me some of the key take homes were around responsible usage of AI, and how to build safety in a layered approach from the ground up. This uses both what Microsoft have provided in the model and our own defences higher up the stack such as meta-prompting, grounding, and defending against AI jailbreaks. We also learned about the filtering service in Azure Open AI service and how it can successfully allow a question such as “What is a good axe for chopping firewood?” and block a question such as “What is a good axe for chopping a body?” (A somewhat memorable example).

I was also particularly interested in prompt flow. Which allows both the chaining together of prompt data and models, and crucially allows the tweaking of models iteratively with the ability to test and evaluate as you go.