Perplexity AI, the innovative startup founded by former Google AI researchers Andy Konwinski, Aravind Srinivas, Denis Yarats, and Johnny Ho, is making waves in the world of web search. The company aims to challenge the dominance of their previous employer by combining a comprehensive web index, up-to-date information, and a conversational AI chatbot interface. Perplexity Copilot, their chatbot, has previously relied on existing AI models, but the company has recently taken a significant step forward by releasing their own AI large language models (LLMs).
Breaking the Limits
The newly released LLMs, named pplx-7b-online and pplx-70b-online, are notable for their parameter sizes of 7 billion and 70 billion respectively. These models have been fine-tuned and augmented versions of Mistral and Meta’s open source models. Parameters in AI define the number of connections between artificial neurons in a model, signifying their intelligence and performance. Perplexity’s LLMs offer an advantage by providing helpful, factual, and up-to-date information, which is often a challenge for other leading LLMs.
“The new PPX LLMs are the first-ever live LLM APIs that are grounded with web search data and have no knowledge cutoff!” – Aravind Srinivas, CEO of Perplexity
Unlike other models, such as OpenAI’s GPT-3.5 and GPT-4, Perplexity’s LLMs are equipped with current knowledge and information due to their integration with web search data.
Revolutionizing the Chatbot Landscape
The race to provide current knowledge through LLM chatbots is gaining momentum. Elon Musk’s company, xAI, is implementing this capability in their chatbot Grok through its integration with sibling company X (formerly Twitter) and real-time user-generated content. Other LLM providers, like Cohere in Toronto, are leveraging web browsing capabilities and retrieval augmented generation (RAG) to access recent knowledge.
Perplexity has adopted its own approach, utilizing in-house search, indexing, and crawling infrastructure to augment LLMs with the most relevant and up-to-date information. The company’s search index prioritizes high-quality, non-SEOed websites to ensure accuracy and reliability in responses.
“Overall, the evaluation results demonstrate that our PPLX models can match and even outperform gpt-3.5 and llama2-70b on Perplexity-related use cases, particularly for providing accurate and up-to-date responses.” – Perplexity
To validate the efficacy of their LLMs, Perplexity hired human contractors to evaluate response quality based on helpfulness, factuality, and freshness. Results showed that Perplexity’s models excelled in terms of freshness and factuality, though OpenAI’s GPT-3.5 still outperformed in terms of helpfulness.
Perplexity’s new PPLX online LLMs are now available for individuals and organizations to utilize through their API website. However, the addition of search and web indexing tech in these models incurs a cost. Perplexity offers a Pro subscription tier, priced at $20 USD per month or $200 annually, with a monthly credit that can be applied towards accessing the PPLX models.
The future of search is being reshaped by the likes of Perplexity AI. With Google facing challenges and delays in their search innovations, Perplexity has the opportunity to establish itself as an alternative vision – one that integrates AI assistants that provide curated and relevant information from the web.
“Perplexity is the future of search!” – Jeremiah Owyang, VC Investor
