Perplexity AI has released two new large language models – pplx-7b-online and pplx-70b-online – designed to provide up-to-date, accurate information by continuously accessing the internet.
These publicly available models address key limitations around outdated responses and hallucinations that exist in offline models.
What exactly is going on here?
Perplexity’s models tap into the wealth of timely data available online in order to produce useful, reality-grounded responses – something offline models still struggle with.
The foundation is built on open-source models like Anthropic’s Constitutional AI, with additional fine-tuning using current web information. This enables the models to better answer questions that require recent information compared to their offline counterparts.
Ultimately, the goal is helpful and truthful responses.
Perplexity’s evaluations show the models perform well in freshness, accuracy, and usefulness – making them well-suited for real-world applications like customer service where up-to-date details are critical.
The launch includes access via public APIs and Perplexity’s AI playground for further testing.
Why this matters…
Consumer AI, like ChatGPT, is beginning to incorporate search, reducing hallucinations and helping satisfy people’s want for real-time info.
However, integrating search is tedious for most developers.
Perplexity’s new online models provide simplified access with SEO expertise already embedded, removing the need to build additional search functionality.
