Always online models from Perplexity

Perplexity has launched two new AI models - pplx-7b-online and pplx-70b-online. These publicly available models will provide helpful, accurate and up-to-date responses, addressing key limitations around freshness and hallucinations.

What’s going on here?

Perplexity’s new models access timely information from the internet to give useful, factual responses - things offline models struggle with.

What does this mean?

The models build on existing open-source models like Mistral-7B and Llama-70B base models. Then these models are layered (fine-tuning) with current web information to answer questions needing recent data better than offline models. The goal fo these models is to give helpful, non-hallucinated responses.

Perplexity evaluated the models for freshness, accuracy and usefulness and they think the models are well-suited for real-world uses like answering consumer questions where up-to-date info is key. As part of the launch, Perplexity has opened up access to the models through its public API and Perplexity Labs playground.

Why should I care?

Consumer products like ChatGPT are adding search to the products. Search results reduce hallucination and people care about real-time info.

But figuring out the search when building with most models is tiresome. PPLX’s new online models API come with their SEO expertise built in for you. No need for that extra search API.