The neural networks become constant assistants of a person in a huge number of areas. More and more users turn to ChatGPT, CLAude, Gemini and other neuralates in search of answers to professional questions or the first information that comes across for student work. However, there is a problem - the so -called hallucinations of artificial intelligence (AI), during which the machine can issue an answer to the question or name the incorrect statement correct. In this material, we will tell you about Perplexity - a neural network that is designed to overcome this challenge.
"Simple" solution for a deep search
Perplexity is a search engine based on the processing of a natural language. Unlike the more popular ChatGPT, Perplexity gives more extensive and understandable opportunities for professionals whose work is related to regular search and clarification of information. She uses the context of user queries to provide a personalized search result on the Internet. This neural network summarizes the search results and gives a text with built -in quotes - which is important, immediately with links to the source of the data. She can also help find more complete information on the topic of interest to you, offering ready -made clarifying requests.
This model in the basic version is able to quickly write a simple working code in the Python programming language for a deeper research or switch to writing mode (disconnected from the global network) and help compose the text. We also note that a number of restrictions that other popular chat boots have is absent from Perplexity.
The tests carried out “checked” show that even in the free version this neural network is quite accurate and unbiased. Apparently, colleagues from Newsguard are in solidarity. After the attempt on Donald Trump, Newsguard Auditwho showed that the products of ten leading companies could not provide accurate information about the arrow and circumstances of the incident. Newsguard checked Chatgpt-4, Smart Assistant, Grok, Pi, Le Chat, Copilot, Meta Ai, Claude, Gemini and Perplexity, using three different requests for six proven attempts about the attempt. On average, chat bots gave a refutation in only 43.33% of cases, and only one could give a refutation to all false narratives, reaching 100 percent accuracy. Newsguard for obvious reasons (the possibility of a lawsuit or accusations of supporting a particular developer) does not disclose which neural network coped, but the independent test “verified” showed that it was Perplexity that is most likely.
As for the search depth, it was Perplexity to the author of this text that helped to find the article in Chicago Tribune, the information from which, in all likelihood, formed the basis of the narrative that The Bible is the most steled book in the world. Other search and AI systems either did not find this link, or hidden enough as an irrelevant request.
Can Perplexity hallucize?
We do not exclude such probability, but in four months of tests we did not observe hallucinations. Of course, this neural network can make mistakes. During testing, Perplexity several times inaccurately read the source and incorrectly compared the data in it. However, this is seen by a small problem: Perplexity provides all links to the texts that I used and easily checked with them. To check with the original sources is a basic skill of factor.
Thin settings
As when working with other neural networks, perplexity can be configured for yourself. In the settings menu, you can give information about yourself and your field of interests, set the style and principles of constructing answers. Having specified as much as possible why you need perplexity, you can get the results even more relevant to your sphere of activity. To supervise the clock for search bias, at any time you can switch to anonymous mode and search without taking into account your settings.
Price and quality
Perplexity works according to the Freemium model. This means that the basic version using its own large language model is available for free, it includes ten premium checks per day. The basic version is programmed for a detailed search, search by scientific data or focusing on a specific source (for example, YouTube). Users of the paid version receive more than 300 premium checks, the ability to choose a language model, including the GPT-4 and the experimental model of their own perplexity engine, access to visualization of requests, images generation and an analysis of an unlimited number of loaded files.
