Exa, a startup, has a different idea: Google for AI. While many startups are trying to replace Google with AI-powered search (we’re looking at you, Perplexity), Exa is different.
The creators of Exa don’t think that humans are the only ones who desperately need a new kind of search engine. Instead, as AI becomes increasingly ingrained in business and consumer life, it will be up to the AI platforms to consistently search the internet for information and provide accurate results—not just delusions. Nor can they use their keyboards to type out their requests.
Exa is developing a tool that will enable AI models to carry out tasks akin to web searches but with native AI experience.
The co-founders invested a million dollars in GPUs, which were more readily available back then. They then developed a machine learning model trained to comprehend links rather than words and sentences natively by using a vector database and embeddings rather than the traditional transformer-based LLM.
Transformers typically anticipate the following word. CEO Will Bryk states, “We train our search engine to predict the next link.” People share links on the Internet, and we use that data to train our model. We also train the model to predict the next link. Therefore, it’s a unique search algorithm.
