Host Intro (Gene Kim)

In previous years, Rosalind has shared her career journey that led her to becoming a technical fellow at IBM. It's the highest rank of individual contributors at IBM, of which there are only 84 active right now.

So Rosalind told me that she recently hosted an all-company watsonx AI hackathon, which included 200,000 people across the entire organization — not just software people, but all roles experimenting. And incidentally, that meant that 200,000 people were hitting internal IBM services all at once.

And so she was here to talk about what the challenges of that were, what the goal was, and how she's bringing AI into the pipeline at IBM. Here's Rosalind.

Rosalind Radcliffe

So thank you. I'm happy to be here again, talking about one of our latest challenges.

For anybody who doesn't know — IBM: software company, hardware company, consulting company. We don't actually do outsourcing anymore, but we have the rest of those parts. We have Red Hat, you know, little company that does a few things. We have over 200,000 people.

And I want to give you a little context. Our CEO loves to do wonderful things for us. And he said we were going to host again an all-company AI challenge. Every single employee should participate. And in his all-employee broadcast, he made it very clear that everyone — it would be good to participate. This was an opportunity to learn, it was an opportunity to grow, it was an opportunity to understand our technology, and it would be really good. It would be a sign of interest in learning.

And there were a few excuses for not participating. On vacation was one of them. Critical assignment on a project. And support — like a few of us who had to support the actual running of the hackathon. Though I did participate.

We had a date that was determined. That was in July. Great time for a hackathon. Isn't that vacation time? Remember, IBM is around the world, and summertime in Europe, that's fun. July 10th through Friday, July 19th was when it was scheduled. And we were all supposed to learn.

This was not just an open challenge — it was a very specific challenge. Each person had to register for the challenge and join a team. Teams could be up to 10 people, no more than 10. And yes, I did get the pings: "Can't I put more people in a team?" And the answer is — and how are you going to work together, collaborate, do this work really efficiently in this one-week time? Well, one week plus 10 people maximum. You could have fewer, but 10 people maximum.

Yes, there were awards. So yes, if you won — and there were a set of winners, three for each challenge, each track — you would get a prize. Now the prize wasn't huge, but you would get a prize. And you would be amazed at the number of people who actually were incented by the small prize. But it was great. And there was a chairman's award, so one of the tracks, one of the key winners would get an additional award.

And these were the tracks that we had. Now if you look at these tracks, some of them you might go, "What's that?" Well, AskIBM was actually a track open to anyone in the company, because that is our internal system that allows you to use gen AI. You can do RAG, you can upload documents, you can ask questions. It's using our internal Granite models to allow you to do what you want to do. And it's already available for all of IBM. So that was one of the options — build a new workflow that helps all of IBM.

We have the Consulting Advantage, which is a set of AI tools that our consultants use. Well, can you make it even better? Can you do something to improve?

You may have heard we have Watson Code Assistant for Z, and we have Ansible Lightspeed available externally. But internally we have WCA@IBM, which supports maybe 120 languages or so as a training base. Would I say it does every one of those languages perfectly? No. But it is — inside IBM, used so that we can test and use it to make it available, and more languages available for our clients. Nice to experiment with a large set of people.

Some of these tracks had a limited number of teams that could sign up for them. And that is one way you can manage to host such a large challenge with so many people — by limiting how many people can access a specific challenge area. And if you look at some of these, you'll notice some of them are more obviously targeted toward developers, and others anyone could participate.

So what happened? We only had 178,000 participate in 128 countries. That was okay — not everybody, but it's a reasonable number.

We had onsite events. There were kickoff events, there were closing events, and there were mentoring events. We provided opportunities for people to come in and see and learn and ask questions, so if they didn't understand they could learn.

We ended up with 11,500 projects submitted. Remember I said they were all judged — sound like fun. They were all judged by two rounds of judging. They were judged by track. And we had a combination of our distinguished engineers and TSMs primarily judging the submissions for the first round.

From the first round, the top 10 of each category went forward to the final round judging. And those of us who volunteered to do final-round judging had to judge many fewer entries. And I did two tracks. It was quite fun to judge. I got to see all the wonderful submissions that got to my round. I didn't have to see all the submissions that got to the first round, which I was very thankful for.

One of the things that was really important was everybody learned. Everybody had an opportunity to get their hands on AI technology. They had the opportunity to grow their skill base with our technology.

Now in IBM, we have a simple rule: if it's IBM's AI, you can use it. Okay? Understand — no, you can't use any of the other choices. You use IBM's. Makes sense, because we build the models, we build the technology. So we are going to use our own as client zero. Now, there are exceptions in areas where we don't work, where we partner with our strategic partners. But primarily when it comes to code, we use our own. Now, our own Granite models are actually open source, so we also open source what we're doing, so anyone and everyone can use it.

But by doing this challenge, we can help improve our models. We had this many people beating up on our system at the same time.

Now, in order to do this, we actually gave each individual team their own cloud account in IBM Cloud, obviously, so they could do the work. However, Watson Code Assistant at IBM is an all-IBM function, and AskIBM is an all-IBM function. And so for those two, we had the opportunity to have a large number of people hit it at once. Networks are fun.

Do you think about what's going to happen when you have a whole bunch of people sitting in a conference room uploading documents to your model, to load into the vector database to be indexed for RAG? Wasn't it fun? So the challenge actually extended into Saturday because of the initial work and a few failures with the network and a few failures with loading RAG documents. It took a little bit. That was the purpose — we did a load test, we broke the system. That was the intention.

Now, what happened? Did we actually do something useful? Absolutely. We generated AskSRE so we can get much more efficient with our SRE practices. Our actual chairman's award winner was, through Consulting Advantage, improving project delivery. We had Watson Code Assistant document an entire large code base. You know, we have some code that might have been written a while ago. We have products that have been around for a very long time. So this was a great opportunity. And as you can see, many others including improving SOX compliance, et cetera.

So it was a challenge and it was a lot of fun.

If you're interested in doing it, we have a model how it works. I doubt many people will see as large a scale as we did, but you can tell our stories. And you can see that we beat up our own products before we make them available, when possible.

So with that, I'm out of time. Thank you very much.