AI Startup Perplexity AI has introduced two new PPLX online models: pplx-7b-online and pplx-70b-online, which can tap into real time information from the internet, enabling them to deliver accurate responses to time sensitive queries. 

For instance, these models can provide up-to-date information on sports scores, stock prices, or the latest developments from Google News.

The models are publicly available via pplx-api, making it a first-of-its-kind API. pplx-7b-online and pplx-70b-online are also accessible via Perplexity Labs, our LLM playground.

The operational framework of Perplexity’s online LLMs involves leveraging open-sourced models, specifically the mistral-7b and llama2-70b base models. Moreover, the process includes utilising in-house search technology and fine-tuning the models to achieve enhanced performance.

The integration of an in-house search infrastructure facilitates the extraction of the most relevant and up-to-date information from the web. The snippets, or website excerpts, obtained through this process play a pivotal role in informing the responses generated by the models.

Initial tests conducted by the company show that Perplexity’s online LLMs not only match but also surpass the capabilities of leading proprietary LLMs, such as GPT-3.5. These assessments cover benchmarks evaluating robustness, helpfulness, and knowledge across various academic subjects.

Traditional LLMs often struggle to provide up-to-date information and are susceptible to generating inaccurate statements, commonly referred to as “hallucinations. The newly introduced pplx-7b-online and pplx-70b-online models aim to overcome these hurdles by not only offering helpful and factual responses but also ensuring the information is current. 

CEO Aravind Srinivas, a former research scientist at OpenAI, co-founded Perplexity with former Meta research scientist Denis Yarats, who now serves as the company’s CTO. The company’s user base has recently reached 15,000 paying customers.