Скачать или смотреть Why AI Costs 100x Too Much but Smaller and Smarter Models solves it - Hans Dekkers from IBM explains

“It’s not only in Asia Pacific, but across IBM globally—we believe, as our CEO often says, that the cost of AI is still far too high.

It’s about 100 times too high. That has to come down. As with every technological innovation, if it becomes cheaper, more people can access and use it. Once it becomes even more affordable, it will naturally integrate into how we operate every day.

One of the main ways to make AI more affordable is through smaller, domain-specific models. These small models are incredibly accurate and can be trained very efficiently.

We don’t need massive amounts of GPUs to train these models—they require fewer resources because they’re smaller and more focused.

At IBM, we believe deeply in the power of small models. There will be a place for large models—probably about 10 to 12 globally—but the vast majority will be small, purpose-built models.” - Hans Dekkers

#enterpriseai #gpus #ibm #aitraining #llms