Perplexity AI, the startup founded by former Google AI researchers Andy Konwinski, Aravind Srinivas, Denis Yarats, and Johnny Ho, is positioned to challenge their former employer’s dominance in web search. By integrating a web index and real-time information with a conversational AI chatbot interface, Perplexity aims to revolutionize the search experience. Their chatbot, Perplexity Copilot, previously relied on established AI models like OpenAI’s GPT-4 and Anthropic’s Claude 2, which subscribers could switch between.
Recently, Perplexity made significant strides towards becoming a premier search destination by introducing their proprietary AI large language models (LLMs) – pplx-7b-online and pplx-70b-online, named after their parameter sizes of 7 billion and 70 billion, respectively. These models, derived from Mistral’s mistral-7b and Meta’s llama2-70b, are fine-tuned versions designed to enhance performance and intelligence.
The uniqueness of Perplexity’s new LLMs lies in their focus on providing current, factual, and relevant information, a feature lacking in other leading LLMs such as OpenAI’s GPT-3.5 and GPT-4. These LLMs are not only accessible through Perplexity’s API for external organizations to build upon but also offer unparalleled real-time data integration.
Unlike competitors like OpenAI and Meta, whose knowledge cutoff dates limit the freshness of information, Perplexity’s new LLMs are designed to provide the most up-to-date responses by leveraging their in-house search, indexing, and crawling infrastructure. This approach ensures that the models are enriched with the latest and most valuable information from the web.
To validate the effectiveness of their new LLMs, Perplexity conducted evaluations comparing responses based on criteria such as helpfulness, factuality, and freshness. The results indicated that Perplexity’s models excel in accuracy and timeliness, outperforming raw models from OpenAI and Meta in certain aspects.
The availability of the PPLX online LLMs through Perplexity’s API signifies a significant milestone, transitioning from beta testing to public access. However, users are required to subscribe to Perplexity’s Pro tier for a monthly fee or an annual subscription to access the models. Additional charges may apply for API calls beyond the allocated credit.
While the adoption of these models by individuals and businesses remains to be seen, Perplexity has already garnered praise from industry experts like venture capitalist Jeremiah Owyang, who envisions Perplexity as the future of search. With the landscape of AI-driven search evolving rapidly, Perplexity’s innovative approach and commitment to real-time information delivery position them as a formidable contender in the field.