From healthcare chatbots to instantaneous marketing materials, generative AI is revolutionizing how we approach business. Monica Livingston leads the AI Center for Excellence at Intel. In this video, she shares her insights into the world of generative AI, how it differs from other forms of AI, and the technology it requires to operate.
Generative AI, as it implies in the name, is actually generating new content from data. So it’s trained on very large data sets, but the output that you get is brand new content. You can even ask it to create content like presentation materials or marketing slides. So in that context, we’ll see generative AI in pretty much every industry.
Hi, my name is Monica Livingston and I lead the AI Center of Excellence at Intel. Generative AI is very, very new, and a lot of the models that we have out there, we’ve not really gotten to the optimization phase. And so, Intel is focusing on that optimization for those models as well to make AI accessible.
For training, we have our Habana Gaudi2 processors, and we have our discrete GPUs, the GPU Max family. And then for inference, for these smaller models, we use Intel Xeon scalable processors, and specifically the fourth-generation Intel Xeon scalable processors because of that advanced matrix extension or AMX engine inside that processor.
And then, we’re spending a lot of time actually optimizing software for our hardware. So for example, we have a tool out of Intel Labs. It’s called SetFit, and what it actually does is enables smaller models to run with the same type of accuracy, but be trained on a much smaller data set. Being able to train on less data, again, makes these models much more accessible.
Most install bases, most data centers have fourth generation Intel Xeon scalable processors in their infrastructure. For us, optimizing these types of models on Xeon means that they are accessible to those enterprises, because you already have this infrastructure, you’re not having to stand up new boxes.
The good thing about AI is that it’s all open source and it’s online. And so, upskilling to AI is really not limited or selected. It’s very accessible. Just think about all of the different types of possibilities. You can already see very context specific models out there for law, for example, for medicine.
That’s extremely helpful to those industries because they’re heavy paperwork focused. And so, when you have to do a lot of paperwork, doing that with some of these large language models saves significant amounts of time.
As generative AI becomes more prevalent and AI in general becomes more prevalent, Intel continues to explore ways to optimize these models and allow customers access to a wide range of hardware and software tools in order to run these models much more accessibly and much more cost effectively.