Perplexity AI 'answer engine'

Perplexity AI is self-described an AI-powered 'answer engine' that aims to deliver precise, contextually relevant, reliable and helpful information in response to user questions.

Launched in 2022, the free version draws on OpenAI's GPT-3.5 and Microsoft's Bing search engine to find relevant information to a question, and uses advanced natural language processing (NLP) techniques to understand and generate an answer.

The paid version, Perplexity Pro, asks the user clarifying questions to refine queries and provides access to GPT-4, Claude 3.5, Mistral Large and Llama 3.

In May 2024, Perplexity launched Pages, which generates a customisable webpage based on user prompts. Pages uses Perplexity’s AI search models to gather information and create a Wikipedia-type page that can be published and shared with others.

Risks and harms 🛑

Perplexity's AI answer engine has been accused of plagiarising and regurgitating third-party content, producing inaccurate, outdated and biased content, and of poor transparency and accountability.

It has also been found to repeat and spread misinformation and disinformation, thereby potentially misleading users, swaying public opinion and undermining political discourse.

Transparency and accountability 🙈

In some ways, Perplexity AI appears transparent - after delivering an answer, the system presents the sources of information it has used, allowing users to verify and explore further, setting it apart from traditional search engines that simply provide links. 

On the other hand, the system is considered opaque in a number of important ways: 

• Data and information sources. Perplexity AI does not reveal all the sources of the information it provides, making it difficult for users to verify the credibility and reliability of responses.

• Training data. The details of the training data and potential biases within Perplexity's AI answer engine are not transparent, raising concerns about the objectivity and fairness of the system's responses.

• Algorithmic processes. Perplexity AI provides limited information on how its algorithms work, notably how they select, filter, and prioritise information.

• Content moderation. The mechanisms and criteria used for content moderation and the prevention of misinformation and disinformation are not clearly outlined. 

• User feedback. The processes by which user feedback is collected, assessed, and integrated into improving the AI's performance are not clearly explained.

Research, advocacy 🧮

• Search Google Scholar

News, commentary, analysis 🗞️

• https://apnews.com/article/perplexity-ai-search-engine-forbes-f307cb607f0db871b05f843a3f744340

• https://www.wired.com/story/perplexity-is-a-bullshit-machine/

• https://musliminsiders.com/here-is-what-to-be-aware-if-you-use-perplexity/

• https://garymarcus.substack.com/p/serious-medical-error-from-perplexitys