Scaling Your Business with AI

Patrick Moorhead:
Welcome back to the Six Five Summit 2024. We are broadcasting live at IBM Think 2024 in Boston. The summit, as you know, is all about AI. In 2023, it was all about the build out, the infrastructure, the tools, and sure as an industry, we’re building those tools out as we speak. But 2024 is the year where enterprises and even consumers are starting to get value. Daniel, how you doing, my friend?

Daniel Newman:
Yeah, Pat, you really hit it on the head. ’23, we saw this big acceleration following the advent of ChatGPT. It wasn’t new. AI wasn’t new, it wasn’t like all of a sudden hearing a thing, but it was that killer workload that everybody could really understand and now it was about how do we extrapolate value from that workload and start to drive to enterprises to drive growth, productivity, and excitement. And that is really what not only all the events we’ve been at this year have been all about, Pat, but what the Six Five Summit this year is all about.

Patrick Moorhead:
And Dan, who was the company that went first to GA with an end to end enterprise AI platform? It was IBM, exactly. And with that, to talk through that, Kareem, from IBM, how are you doing, my friend?

Kareem Yusuf, Ph.D:
I’m great guys, thanks for having me. It’s great to be here

Patrick Moorhead:
Back for a second year at the summit. We are so excited. And by the way, are you just going to just do a victory lap the whole time here?

Kareem Yusuf, Ph.D:
No, not quite. We’ve got a lot of work to do, but it’s good to be making progress. It’s good to be seeing some traction, so yeah.

Patrick Moorhead:
For sure.

Daniel Newman:
Yeah, so we sat down a year ago, actually the studio wasn’t quite this cool, but we had a great conversation actually for our Six Five Summit.

Kareem Yusuf, Ph.D:
That’s right.

Daniel Newman:
And we’re bringing you back again because it’s been a massive year. In fact, in the AMA, we spoke to Arvind, I hit them up on the year in review, the 4.4 trillion of productivity that he keeps talking about. But you’ve got growth, you’ve got product, you’ve got the responsibility to bring this story to life for IBM and its customers. Talk a little bit about how that’s moved throughout the last 12 months.

Kareem Yusuf, Ph.D:
Well, it really anchors in first and foremost on what I would call use cases. So the whole journey began, as you recall when we were talking last year about how was GenAI going to be brought for business to the enterprise? And as you looked over the year that went past, I would say three things very clearly emerged, customer service as a great starting point for thinking about bringing GenAI to bear was emerged. This notion of digital labor tied in with business process alternation, integrating GenAI to actually doing work was the next. And then a little unexpected for me only because I thought developers would be like, “Stay away from me with all this AI.” GenAI for code development really exploded as a real way to unlock productivity benefit. And so, that really shaped the evolution of the last year and the traction we’ve been seeing.

Patrick Moorhead:
So a question, you talked a little bit about the use cases. Can we do the double click on that and maybe talk about some of the outcomes that people like to see? And if you want to cite outcomes that your clients have had, that would be great too.

Kareem Yusuf, Ph.D:
Well, I mean, look, if you think about outcomes, there’s the reason why everybody talks about productivity is because it’s very easy to begin to think in terms of time saved and then based upon time saved, things either more done or avoided. So let’s take customer service as an example. A lot of the business cases center on the ability to serve without having to escalate to a human, that has got very clear benefits in terms of time computation. Also, when you think more along that notions of avoidance, think about an AXE HR system. Self-service is very empowering and the more you can enable people to do stuff themselves, the more you drive that kind of value. We saw it extensively at IBM. We saw it at many of the customers who you saw on stage talking over the last three days. Elevance Health as an example.

Dun & Bradstreet was just another one just thinking immediately off the top of my head. When you think about code as another case, it speaks to this notion of productivity, but also expertise and skills enablement. We introduced the product, watsonx Code Assistant for Z, for example. A lot of our customers love using GenAI there to understand the COBOL applications. Just being able to gain that understanding and bring that kind of knowledge to the fore before you take action becomes valuable. So that’s kind of how it’s beginning to manifest in real tangible terms.

Daniel Newman:
That’s great. So I mentioned in the preamble here about ChatGPT, but let’s face it, we’ve moved well beyond that being the only model, the only consideration, it set the stage, it got people going, but now we’re seeing private AI proliferate. We’re seeing hybrid architecture, stuff that actually you and IBM bet on very early on. We’re also seeing this kind of range of model sizes. We’ve got challenges with power, we’ve got challenges with total amounts of data that are available to be used. What are you sort of seeing as it relates to the model development and data used to create them?

Kareem Yusuf, Ph.D:
Yeah. Well, look, as you well pointed out, our thesis was proven out that targeted models would begin to rule the day. And also that price performance of models in terms of inferencing costs and all that would become more and more critically important as people looked at the size of models that they needed to deploy. But I think what really became the key unlock from all of that was this focus on how do models get better and how can you source more contributions to improve a model?

And that led us to the InstructLab technique, which we then open sourced via Red Hat along with key models that we have our granite models, one for English language-based LLM, and then a whole host for code-based models to really begin to solicit that open source involvement. And I want to stress a very important point when I talk about that ability to contribute. Before InstructLab was put out there, the state of the art was fork a model and do something with it. And so, you take for example, a Llama model and you’ve got a gazillion variations of Llama, but no way to bring all of that together into a single model.

Daniel Newman:
And when you say fork just for everyone out there, you basically mean like a save as in a document.

Kareem Yusuf, Ph.D:
Thank you, that’s the exact point. That means save as. So you take the one save as make changes. And so, you’ve got all these copies in a-

Daniel Newman:
No version control, no ISO.

Kareem Yusuf, Ph.D:
No version control, no nothing.

Daniel Newman:
No TQM, nothing.

Kareem Yusuf, Ph.D:
But the InstructLab technique says you can actually bring together various contributions around skills and data, enhance the model. And very importantly, when you think about enterprise customers in the way we’ve licensed this, enhance it for private use or enhance the contribute back into the public domain, I think this will unlock a lot of innovation as we go through the next coming year.

Patrick Moorhead:
Yes, analysts, we’re still kind of peeling the onion back on InstructLab, but it seems first of all very provocative and from what I understand aligned with a lot of the needs of the enterprise that we talk to who they use a large language model, a big one, and they’re not getting the results even through, let’s say utilizing RAG and then they feel like they’re in the position where they’re like, “Hey, I need to create my own proprietary model myself.” And they look at the cost and the ability to do that. So this is a really interesting way to kind of have your cake, you eat it too. So I gave you kudos up front about WatsonX, first enterprise platform that went GA. Can you talk us through, it’s been GA since July, how has it evolved since then?

Kareem Yusuf, Ph.D:
So I think if you remember when we GA’d the platform, there were three core components or our three core components, .ai, kind of the studio environment, .data, the data lake house and .governance, which was for doing governance. All of them have come through a number of terms of the crank over the course of the year with some key announcements being made this year. As we were at Think and as we look ahead. On the .ai side, it’s been all about how do we make it easier for these model outputs to be embedded into applications. We already started with this notion of providing multi-choice, access to a lot of different open source models and ours. And that has continued with multi-language models coming in, dedicated models for Arabic, for example, the alarm model coming in and obviously as we were talking about open source models and the like. But the next level is when you look at how people are building applications that leverage these models, there’s a lot more that goes into it, a lot more considerations and how do you do that at scale?

So looking at all the various frameworks and how do you bring that together in an enterprise context? That’s been a key element there on the tooling side. On .data, it’s all about performance. It really was an open data store, open fabrics, but it was all about open formats rather how do we up the performance, bring it in the new Presto engine and really giving a price performance element that really sinks. And then doing some of what I call recursive and embed GenAI there to allow the data and the metadata to be better understood for those working with that. And then last but not the least on governance, we already had what we’re doing with monitoring and all of that, and we had started down the path of processes with taking things like various regulations and creating regulatory packs. But the next key unlock was opening that up to support models regardless of where they run. So whether they’re running on AWS SageMaker like we demonstrated this week and showed or anywhere, we can now use Watsonx.governance to bring that kind of governance framework to it to bear as well.

Patrick Moorhead:
That’s great.

Daniel Newman:
And there’s going to be a ton of focus on governance, tons of focus on compliance and safety of AI in the coming months. I recently did a segment, we were talking about privacy and someone’s voice was being used, we’ve got licensing issues in use and every enterprise is creating this new sort of immeasurable risk when they’re trying to move this quickly, but not being quite sure how models are trained. And so, this governance component is incredibly important. Another thing that IBM has been very focused on, Kareem, has been around these alliances. You do the AI Alliance, you’re working closely with companies like Meta, who’s been very focused on open source. Of course, IBM owns Red Hat, but Red Hat operates very much in parallel and has a massive developer in open source community. This has been an enabler of so much progress for you, but can you talk a little bit about how ecosystem, how partnerships and these alliances are really shifting and how it’s worked over the past year?

Kareem Yusuf, Ph.D:
Well, look, as you know, ecosystem has always been important to us and this kind of open mentality, and it’s really about bringing together sets of folks who want to collaborate together around ideas and moving forward the state of the art. You’ve heard me say before, technology is a means to an end. There’s an end our clients are actually trying to achieve in terms of their businesses and what they’re trying to do. So yes, you mentioned correctly, strong relationships with Meta, strong relationships with Mistral as well as many others opening up to various regional concerns as well. But take for example, your point around Red Hat, that’s also about how do you lower the bar of entry and hit those hardcore developers who are operating at that level.

That’s why we introduced Real AI, which really is on the back of Red Hat Linux in the InstructLab and what you might call it, open models so that you can get very small developer laptop footprints that folks can start working on, that allows us to create this value chain from the very beginning there all the way up to enterprise level tools with more in them with what’s next. So it’s really critical. The last element I’ll say on ecosystem, not to be forgotten, is also remember that we’ve been working with so many ecosystem partners to embed what’s next in their solutions. We talked about Adobe, obviously we talked about SAP before. Airtable was on stage earlier this week, right? We’ve been talking about all these folks who are helping us to get better and really generate this ecosystem of innovation to ensure we’re delivering real value that matters.

Patrick Moorhead:
On the ecosystem side, I have to admit, so I’ve been tracking IBM for almost 35 years and I feel like this is a great example, the ecosystem play of what I consider the new IBM. And it’s beneficial because addressing clients’ needs really takes a village. Not one single company-

Kareem Yusuf, Ph.D:
Exactly.

Patrick Moorhead:
… can do it either horizontally, vertically in a certain country, and it’s just great to see you do this. So second year in the Six Five Summit, we really appreciate your participation. We’ve talked a little bit about what you’ve done since May. We’ve talked about the updates that you’ve had since then, talked a little bit about what you announced at IBM Think. Let’s talk about the future a little bit. Where does this go? Next year when hopefully graciously come on the show again and represent the summit in ’25, what do you want to have accomplished?

Kareem Yusuf, Ph.D:
Look, I think of three lanes of activity. In the context of the core platform, it’s continuing to do the work that helps to accelerate leveraging large language models within the applications, within agents, bringing that more and more to the enterprise. And I think that’s a theme that we’ve got quite a bit of way to run with to strengthen. The middle lane for me is all about the notion of AI assistance and making it easier to build AI assistance, to integrate AI assistance that are tailored, automate critical process and integrate into the enterprise. Our lead offering for that is the watsonx Orchestrate offering in terms of that build, but also then building and delivering high value assistance to target very specific domains. You’ll still see more on that theme. You saw, as I said, the watsonx Assistant for Z, watsonx Code Assistant and that family expanding becomes critical. We just brought in yet another member of that family around enterprise Java applications.

So you’ll see a lot more activity on the assistance. And then the last lane, not the least, but not to be missed, the ongoing embedding of these AI features into what I would call products that people just use from us in various domains. Think about what we showed, watsonx Assistant embedded into planning analytics or Aptio or Maximo or Guardian continue to strengthen and broaden that out. And that’s really important because for many customers, that will also be their first taste of GenAI at scale as they’re interacting with these products and getting differentiated value from us. And that’s why we have to be really targeted there to make sure it’s not just, “Hey, here’s another chat board for help.” It is embedding assistance that brings real value to the work they’re doing in those applications.

Patrick Moorhead:
It’s exciting. A lot of work cut out for you this year in your team.

Kareem Yusuf, Ph.D:
A lot of work cut out for us and we’re really looking forward to it. And a final footnote on that, this all leads to when you think about that last year of applications, keep paying attention to what we’re doing in AI enabled automation, especially around IT operations. We launched IBM Concert, which is purely a GenAI native application that is able to synthesize data from the entire IT environment, gives a real view on what’s going on with applications, and drive meaningful value on how they get managed. More of that too.

Daniel Newman:
You want it to be scalable, you want it to be turnkey, you want it to be available, you want it to be responsible. Next year, you’ll bring a star to the show. Kareem, thank you so much for joining us here on the Six Five. We appreciate you joining the summit this year.

Kareem Yusuf, Ph.D:
Thank you for having me. Cheers.

Patrick Moorhead:
Thank you.

Daniel Newman:
All right everybody, we are here at the Six Five Summit 2024, brought to you here at IBM Think 2024. We appreciate you tuning in. Stay with us for all our coverage. We’ll see you soon.