I had the pleasure of speaking at the CuttingEdgeAI event organized by NORA today. I told my story through the lenses of Brua the Critical.
Brua the Critical
Brua the Critical is a research chatbot that grounds conversations in reality. Under the hood, it integrates multi billion parameter (large) language models (LLMs) from Cohere and an expanding suite of reliability capabilities. As I told my story, we examined the techniques for reliable training and deployment of LLMs that AI systems like Brua the Critical utilise. Along the way, we saw the good, the bad and ugly, and the useful side of LLMs.
But what is the story?
I was leading the large model training stability program at Graphcore, an AI chip and systems company, up until 2022. We delivered software that made life easier for those who want to (pre-)train large models end to end on multi IPU systems without suffering numerical collapses that are otherwise rife in the process for all manner of AI hardware. Users would enable this functionality with a single line of code. I am proud of this and the team that made it happen.
While I am no longer at Graphcore, I carry its ambition with me — in fact I carry it literally in my pocket (a souvenir IPU chip). Now, I built my first robot from scratch at home in India in the year 2000 together with a couple of great engineers in the making. I wrote my first neural net from scratch in 2003 in the UK, where I spent a good number of years before moving to Norway as an academic in 2011. I have explored academia, lived the life of an entrepreneur, and have ridden the deep learning wave in the industry.
After roughly 20 years working in the field, hindsight, the present state of affairs, and the future of my kids growing up with AI capabilities as first cultural concerns, I have converged on what I want to do next, and to do so in Norway, despite the misguided narrative that one has to move to the Bay Area to make a dent in the universe.
I consider Norway to be the epitome of human values. I have also directly worked with people here who have built AI systems that supersede those built by Deepmind, OpenAI, Google and NVIDIA. There is no dearth of talent and vision.
My ambition is to give the world an alternative path in the ongoing AI arms race, a path with human values. With brua.io, we aim to transform the way humans interact with AI capabilities. We are here for these capabilities to augment human endeavours reliably and safely. We are also here to build a deep tech company rooted in Oslo that expands the frontier of AI along the way.
We strive for a culture where humans and AI work alongside each other, and do so with the same trust that we have in a bridge, or brua if you like in Norwegian, as we walk over it.
If you are interested in furthering this ambition, let’s talk, and let’s shape the future of AI right here!
This is my story, and I am on page 1.