We're witnessing a transition in AI. Narrow systems able to execute specific tasks in a single domain are giving way to broad AI that learns more generally and can work across domains and problems. This shift is being bolstered by foundation models, which are trained on large, unlabeled datasets and fine-tuned for an array of applications. They are transforming how we compute at scale. But we need more innovation in hardware and infrastructure to power and deploy this next generation of AI models. Continuing to rely on conventional infrastructure and processors is simply not scalable — or sustainable.
By invitation only, over 240 visitors attended from 27 companies and government agencies and 14 universities. Held at the T.J. Watson Research Center in NY, guests discussed what's next in foundation models and the challenges in designing AI hardware to support complex and multi-modal workloads. The forum included presentations from industry leaders, interactive demos by IBM experts, and a poster session showcasing research from leading universities.
Role: Design lead
Photos: Craig Warga
Designed at IBM Research
Designs were based on images of an AI hardware wafer we photographed under a microscope.
Since chips are inherently square, our concept was to find a way to represent them in an opposite way. To me, one of the furthest forms from square is the movement of clouds, coming together and apart again. This idea was conveyed much more successfully in motion than static applications, though the organic shapes still worked.
The level of detail can change depending on scale and use. How the elements lock up is also flexible. Depending on the application, we use the entire lock up, square and chip or just the wordmark.