New Perplexity Feature Changes AI Search Forever (Chain of Thought Prompting)
Perplexity is a powerful AI search engine that can be used for a number of different things - this video is supposed to show you some of the best features of this incredibly powerful tool, as well as showing exactly how I use the tool
The first method is using its ability to agentically search the internet
Give it a data prompt to find data
can you find real data that could accompany this article - Do not add the data to the article, just give me interesting data as a summary. Give historical data where possible so it can be used as a comparison
Then watch exactly how it does the research - it’s very interesting how it does it. First it discusses the question, then it forms the question, then it does several searchers of the same question
This is agentic in its nature because it doesn’t just do the search or make up the data straight away, it first reasons the question out - which is a huge feature of AI that you would normally have to do some fairly complicated programming for
What this means is that the average answer is much more accurate to what you’re asking because Perplexity AI Pro has a reasoning step. This reasoning step is key in getting good results
This is further proven by the API vs using Pro. If you don’t know - Perplexity API uses Llama 3.1 on the API and performs significantly worse than the pro version of perplexity - actually I’m fairly close to replacing Perplexity with my own scraping system purely because i’m not massively impressed with the perplexity API - That is the real proof that the power is in the frontend of Perplexity, not the API.
The is particularly powerful because of how powerful agentic workflows actually are
It allows AI to mimic reasoning, which can be defined as understanding problems (before tackling them).
You can think of Perplexity itself as one agent which formulates questions for the searcher agent to answer, and then the scraping agent scrapes key details from the search results
If you look in detail at how to actual search of perplexity is done, it’s actually a fairly simple process
Read the prompt from the user
Generate 2-3 steps to complete the task
Generate questions or research certain topics in each task, doing so at several points
Answer each question in each part one by one
Consolidate all of the information with a consolidation prompt
This is called Fusion Chain prompting, and is a huge improvement on the original concept behind Perplexity, which was to do a fairly useless Google search and then scrape the results and try to answer the question based off of these results
The fusion chain prompting method gives it a huge accuracy boost and gives the results a much needed boost in quality, which is why I recommend using Perplexity pro for research steps - unless you have an entire system like Harbor to do all of this for you (which is what we do)
Try our SEO tool: https://harborseo.ai/
Work with us: https://incomestreamsurfers.com/