Perplexity Debuts 2 Online LLMs with With Real-Time Knowledge
Generative AI startup Perplexity AI has introduced two new large language models (LLMs) that offer real-time access to the internet and information online. The pplx-7b-online and pplx-70b-online models are designed to overcome two significant challenges commonly faced by traditional LLMs: staying current with fresh information and avoiding the generation of hallucinations and inaccurate statements.
Accessible through the pplx-api and Perplexity AI’s LLM playground, these models represent a novel approach in the realm of AI-driven information retrieval. The key feature of pplx-7b-online and pplx-70b-online is their ability to access and utilize information from the internet, allowing them to provide responses that are not only helpful and factual but also up-to-date. The development of these models builds on the foundations of existing open-sourced models, namely French startup Mistral’s Mistral 7B and Meta’s Llama 2 70B. The new models build on top of existing base LLMs but additionally connect to Perplexity’s proprietary search infrastructure, spanning millions of web pages. The system extracts text and data to augment the models with the latest information for time-sensitive queries.
“By providing our LLMs with knowledge from the web, our models accurately respond to time sensitive queries, unlocking knowledge beyond its training corpus. This means Perplexity’s online LLMs can answer queries like “What was the Warriors game score last night?” that are challenging for offline models,” the company explained in a blog post. “Perplexity’s mission is to build the world’s best answer engine – one that people trust to discover and expand their knowledge. To achieve this, we are deeply focused on providing helpful, factual, and up-to-date information.”
Perplexity tested the new models with crowd-sourced test questions aiming to stress shortfalls in each area and had human evaluators compare outputs from the pplx-online models versus alternatives on which response better met the criteria. The pplx-online models strongly outperformed across categories in the early results, demonstrating real progress on prevailing LLM pitfalls, according to the company.
The new models continue Perplexity’s ongoing expansion of its portfolio, including its mobile app and browser extensions. Both added the ability to answer questions with images over the summer, and Perplexity was among the first to launch a chatbot built on Meta’s Llama 2 model. The startup also offers a copilot feature for personalizing answers, which is unrelated to the Microsoft tool. Perplexity fueled the new features with revenue from its conversational search engine apps, as well as a $25.6 million funding round in March.