Perplexity AI Launches Online LLMs to Challenge Google Search Dominance

Perplexity AI, a startup founded by former Google AI researchers, has unveiled its own large language models (LLMs) in an ambitious move to challenge the long-standing dominance of Google in web search. The company's new models, pplx-7b-online and pplx-70b-online, are designed to provide up-to-date, factual information through a conversational chatbot interface, setting them apart from existing LLMs like OpenAI's GPT-4 and Anthropic's Claude 2.

Why it Matters

Perplexity AI's new models are significant for several reasons. Firstly, they integrate web search data in real-time, ensuring that users receive the most current information available. This is a critical feature that many leading LLMs, including GPT-3.5 and GPT-4, lack due to their knowledge cutoff dates, which were limited to September 2021. Secondly, Perplexity AI's models are available through an API, allowing other organizations to build applications on top of them, thereby expanding the potential use cases and reach of these LLMs.

Key Risks

Despite the promising features of pplx-7b-online and pplx-70b-online, there are several risks associated with Perplexity AI's venture. The primary challenge is competing with Google, a behemoth in the search industry with vast resources and a well-established user base. Additionally, ensuring that the models remain up-to-date and accurate will require continuous investment in data collection and model training, which can be resource-intensive.

Another risk is the potential for misinformation or biased information, as integrating real-time web data can introduce inaccuracies. Perplexity AI will need to implement robust mechanisms to filter and verify the information its models provide to maintain user trust.

The Opportunity

The opportunity for Perplexity AI is significant. By offering a search experience that combines the conversational ease of chatbots with the accuracy and currency of web search, the company can attract users who are frustrated with the limitations of existing LLMs. The availability of these models through an API also opens up new possibilities for developers and businesses looking to integrate advanced AI capabilities into their products.

Moreover, Perplexity AI's approach aligns with a growing trend in AI, where models are not only intelligent but also contextually aware and continuously learning. This can lead to more dynamic and interactive user experiences, potentially disrupting traditional search paradigms.

Technical Details

The pplx-7b-online and pplx-70b-online models are fine-tuned versions of the open-source Mistral 7B and LLaMA2 70B models, respectively. The parameter sizes-7 billion and 70 billion parameters-indicate the complexity and potential capabilities of these models. Parameters in AI refer to the connections between artificial neurons, with higher numbers generally indicating more powerful and knowledgeable models.

Perplexity CEO Aravind Srinivas highlighted the unique features of the new LLMs on X (formerly Twitter), stating that they are "the first-ever live LLM APIs that are grounded with web search data and have no knowledge cutoff!" This real-time integration of web data is a key differentiator, as it allows the models to provide relevant and current information.

Conclusion

Perplexity AI's launch of pplx-7b-online and pplx-70b-online represents a significant step in the evolution of search technology. By combining conversational interfaces with up-to-date information, these models have the potential to challenge Google's dominance and open new avenues for developers and businesses. However, the company will need to navigate the risks associated with maintaining accuracy and competing in a highly saturated market.