Photo by Om siva Prakash on Unsplash

Perplexity; combining the strengths of generative AI and traditional search

Emil Rijcken
Geek Culture
Published in
3 min readFeb 17, 2023

--

In Late November 2022, ChatGPT was released by OpenAI. It directly broke many records. After four days, it had more than one million active users, which rose to 57 million after a month. Before ChatGPT, Google+ held the record for reaching 100 million users the quickest. They did it in one year and two months. However, only two months after its introduction, ChatGPT’s user base has already surpassed 100 million in January 2023.

Obviously, these rapid developments were closely followed by Google. Getting a question answered directly is much easier than looking for it through a list of URLs provided by a search engine. In December, Google expressed alarm internally to address the risk of unexpected strength of ChatGPT to its search engine business, which accounts for 71% of its total revenue.

Luckily for Google, ChatGPT will unlikely take over the search business anytime soon. Although its ease of use and answers are impressive, it has some significant downsides too. It is trained on (lots of) data until 2021. Hence, asking questions about recent events will likely return an answer indicating ChatGPT’s inability to answer. Furthermore, there is no guarantee that ChatGPT’s answers are valid. Being a large language model, it is trained to maximize the likelihood of a sequence of words given an input query*. However, there is no truth validation. Meanwhile, its answers are returned convincingly, easily tricking users into believing what they read. Furthermore, ChatGPT does not show the sources it used to answer questions so users cannot even validate the answers themselves.

Based on these limitations, Google might have given itself time to breathe in December. Then, another tool was released: Perplexity.AI.

This search engine returns a story-based response to your query, similar to ChatGPT. However, instead of leaving the user guessing whether the returned answer is truthful, it uses citations to show the information sources. Less than 24 hours before writing this story, the Superbowl 2023 was played. When asking who won, the answer below was given.

Hence, ChatGPTs time limitation does not apply to Perplexity.AI. Other useful features of Perplexity.AI are its suggested related search queries and the possibility to ask follow-up questions. Unfortunately, no paper has been released for Perplexity.AI either. However, likely the follow-up answer considers the previously asked question(s) as well.

So why then still use ChatGPT?

Currently, Perplexity.AI only allows for a query length of 255 characters at most. Also, it doesn’t come close to ChatGPTs creative features.

Where Perplexity.AI gives a short answer, ChatGPT is much more elaborate:

Also ChatGPTs coding skills are unmatched

The answer provided by ChatGPT is incorrect and its layout is less appealing than the ChatGPTs correct answer.

Hence, when programming or working on creative work, ChatGPT is your go-to tool. Whenever seeking the truth intuitively, Perplexity.AI is better.

*OpenAI has not (yet) released a paper describing the details of ChatGPT.

--

--