Improving Radar Performance for Physical AI Systems

Over the past several years, the term artificial intelligence (AI) has become largely associated with large language models (LLMs) like OpenAI’s ChatGPT and Google’s Gemini. LLMs are trained by feeding massive datasets into transformer neural networks that convert pieces of information into tokens. The model then analyzes patterns, specifically, the probabilities that certain tokens will follow others, to generate output.

LLMs initially gained widespread popularity as chatbots that respond to typed or spoken prompts. However, they represent only one category of AI models being used to try to replicate and supercharge the human brain’s functionality.

One concept that has emerged from recent LLM research but is broadly applicable across other AI approaches is the reasoning model. At their core, reasoning models break down a request into smaller components. Each component can then be processed by the same LLM, different LLMs or other AI models. As each component produces an output, so-called “thinking tokens” pass information along a chain of thought that guides subsequent decisions. The models used for processing can be selected based on the task requirements.