A few days ago, a piece of financing news reached China. Exa, a laboratory located in San Francisco, California, announced that it had raised $22 million. This funding was led by Lightspeed Venture Partners, with participation from NVIDIA's venture capital arm and Y Combinator. Exa's goal is to create a brand new search engine specifically designed for artificial intelligence.
Exa's founders are very young, with CEO Will Bryk at 27 years old and co-founder Jeff Wang at 26. Notably, they founded the company before ChatGPT was launched.
Exa initially built a tool that allows AI models to perform operations similar to web searches. This includes finding information from the internet, AI chatbots that help answer customer questions, and providing training data for companies.
The founders initially invested $1 million in GPUs, using vector databases and embedding technology to build machine learning models. The model was trained to intuitively understand links rather than individual words or sentences.
Will Bryk explained that their search engine differs from ordinary search engines in that it doesn't guess the next word, but predicts the next URL a user might click. This approach trains the search engine based on links people share online, representing a completely new way of searching.
After ChatGPT's explosive popularity, many AI companies began requesting API versions of Exa's search engine to integrate into their own models. Currently, thousands of developers are using Exa's products, and the customer base is continuously growing.
Exa's founders are dissatisfied with the current internet environment. They believe that the internet was originally a place to easily access information, but has now become increasingly commercialized and distorted due to the competition for attention. Especially in Google searches, the existence of the search engine optimization (SEO) industry means that search results may not provide the most useful information.
Exa's search results are displayed differently from traditional search engines. It offers various filtering options such as PDFs, GitHub, companies, news, print media, tweets, podcast posts, etc. Users can choose different information sources according to their needs.
Exa also provides domain filters and phrase filters, which can improve the accuracy and efficiency of searches. However, compared to Google or Perplexity, Exa's user experience is more technically oriented and may not be as suitable for average users.
In contrast, Perplexity is a search engine designed for general users, offering a question-answering search experience. Exa is primarily designed for AI systems and developers, providing the knowledge and data needed for AI.
Exa uses Embedding technology to understand semantics and can search various data sources such as Twitter, GitHub, and Reddit. Perplexity, on the other hand, uses GPT-4o API and language models like Claude-3 and Sonar Large (LLaMa 3).
Embedding technology can convert textual information into numerical vectors, allowing machines to "understand" and distinguish different concepts. GPT-4o API and Claude-3 are pre-developed language models, and Perplexity integrates different models for tasks such as writing articles, answering questions, and chatting.