• by Laurie Sullivan  @lauriesullivan, Yesterday

Advertisers will soon have faster hardware and software support to run all types of features being built in artificial intelligence.

Google Ironwood, the company's seventh-generation Tensor Processing Unit (TPU), and the first designed for large-scale AI inference, will launch later this year. The AI accelerator chip optimizes hardware and software for advanced AI workloads, including those in Gemini models.

The company believes Ironwood will significantly impact advertising by enabling advertisers to improve campaign performance based on increased processing power and more efficient execution of AI models.

It will generate personal experiences, advance ad creation, enhance and increase audience targeting through accelerated processing of user data and behavioral patterns. It also will give advertisers the ability to explore new advertising experiences and ad formats.

Inference is a major part of what makes AI more capable to make all types of calculations and decision.

Deep-learning AI language models, image generation models and audio models all use inference because they make predictions for what will happen next based on what they’ve learned from past data patterns. Recommendation models use inference, too.

“Most ads models are recommendation models, and the model that recommends YouTube videos to you,” Fenghui Zhang, Google senior product manager, wrote in a post.

Models today are better at physics and texture, among other things like text translation, Zhang wrote.

For example, language translation used to be statistical. It was usable but it wasn’t correct or conversational. Statistical translation led Google to generative AI (GAI) translation, which lots of people today feel comfortable using, even in customer-facing products, Zhang wrote.

“We’re still using the process called inference, but the underlying AI and our computation capacity have improved dramatically,” he wrote.

Inference is not just what allows AI models to predict, but also allows them to classify context, Zhang wrote, adding that models can label things based on how they have learned.

Niranjan Hira, Google distinguished engineer, referred to inference as pattern matching. When discussing inference and generative AI, it’s as if asking if AI models match patterns to predict what you want.

“For example, if I said, ‘peanut butter and ____,’ and asked an American audience to fill in the blank, they’d probably say ‘jelly.’” Hira wrote. “That's a good example of inference for speech patterns, and that’s something AI inference can do, but it goes way beyond that.”