Perplexity and ChatGPT are two well-known artificial intelligence chatbots. They were released to the public during similar timeframes, and they both address similar problems that their approach to AI is supposed to help solve.
When it comes to research, writing assistance, and similar tasks, these two chatbots deserve a closer look. That way, you can make a more informed decision about which is more helpful.
What Is ChatGPT?
ChatGPT is an AI chatbot created by OpenAI. It’s probably the one you know the most about if you’ve been following the news. When ChatGPT (GPT-3) was released to the public in 2022, AI held a new place in public discourse and consciousness.
OpenAI has been working on generative AI since 2015. Now, anyone can create an account and interact with the most advanced AI chatbot technology. It will quickly be used (and misused) for numerous academic, business, work, and entertainment purposes.
What ChatGPT does on the surface is simple but novel and impressive. ChatGPT can understand and engage with users with human-like text. There have been plenty of chatbots before. But now ChatGPT gives people an experience of AI that can:
Understand human languages
Learn context and nuance
Formulate opinions and arguments based on billions of data points
Complete a wide range of text- and code-based tasks
Hold exciting and in-depth conversations
What set ChatGPT apart from the start was its knowledge range and ability to understand and communicate. Even now, GPT-4 and GPT-4o have the largest parameter counts and perform at the highest levels on standard AI benchmarks.
Since the public release of ChatGPT, OpenAI has remained ahead of the curve with its new models. But while it’s been an impressive tool, many other models have also emerged. In many cases, former OpenAI employees and even founders have gone on to start their own alternatives. But it isn’t for nothing that many of those involved in Claude, Grok, and Gemini were formerly working at OpenAI.
Given that ChatGPT set many of the AI precedents in motion, it’s essential to first look into what it is.
How ChatGPT Works
ChatGPT is an AI large language model (LLM) trained to understand and generate text based on user inputs. The platform generates text with a transformer-based architecture, which is the chatbot’s namesake, the generative pre-trained transformer (GPT) architecture.
The OpenAI GPT was trained on masses of data from millions of sources. GPT training materials include:
Textbooks
Website content
Academic studies
Other text sources
During the training, ChatGPT was taught to understand even intricate language features. It can understand patterns, grammar, logic, reasoning, and nuance. The conversational context was built into the training, making ChatGPT carry human conversations while remembering everything and expanding the conversation.
The “transformer” in GPT processes language by breaking text down into smaller tokens. Transformers can understand the relationships between each letter and between each word. This is why if you use ChatGPT, you’ll normally get complex but coherent responses.
Every interaction with ChatGPT feeds in additional tokens, adding layers of mathematical reasoning and pattern recognition. Building from that context, ChatGPT puts its training-provided knowledge to work. Mathematical operations do the rest of the work, and the GPT creates a response. Overall, the natural language processing (NLP) entails:
Tokenization
Semantic analysis
Entity recognition
Sentiment analysis
Machine translation
Text generation
Simply put, what ChatGPT is doing when given input is guessing what should come next. It’s a very mathematical operation. Despite how coherent its responses are, ChatGPT doesn’t “think” in a way that humans do. It relies on patterns and probabilities. This places many limitations on ChatGPT, but also makes it very useful for language-based tasks for any subject. That’s why many users go to ChatGPT for everything from casual conversations to professional assistance roles.
Apart from lacking human reasoning, ChatGPT’s main limitation is its reliance on training data, or “parameters.” The most recent version of ChatGPT-4 contains over 1.8 trillion parameters.
Multimodal
ChatGPT is capable of multimodal input, meaning going from one content form to another. The most well-known example is that you can give it text prompts, which will produce an image for you. Likewise, you can generate responses using text or audio.
Custom GPTs
Some users replace the ChatGPT plugin system with customized GPT functionality. You can find or even create unique versions of ChatGPT that suit your needs. You don’t need coding abilities to get highly specialized use out of ChatGPT.
What is Perplexity?
Perplexity AI is a search engine that uses AI technology to answer inputs using live information on the web. Similarly to ChatGPT, all the underlying technology that goes into ChatGPT is hard at work. But in some cases, the response it gives is literally provided by a GPT model with internet access.
Perplexity was launched in December 2022 by four founders who met while working at Google AI. They used their LLM experience to handle one big problem that other companies have only recently been addressing: fully unlocking the knowledge in AI models.
The name “Perplexity” nods its head to the company’s mission of giving users more recent and relevant information. They wanted to provide complete and well-researched answers to even the most difficult questions. They also wanted to “democratize” knowledge and AI by using it to act as every user’s research assistant.
Unique Perplexity Features
The Perplexity interface looks a lot like ChatGPT’s, with a few noticeable additions. Whenever you ask the chatbot a question, it enumerates its answers, corresponding to citations at the top of the page. The citation system vastly speeds up Perplexity users’ experience with the platform as a research tool.
Citations From the Web
One of the main strengths of Perplexity is that it expedites research. When you ask Perplexity a question, it quickly applies the AI model of your choice to search the web. The answer it gives you will include citations and direct links to the source of information.
Like other chatbots, Perplexity doesn’t always provide authoritative or factual information. However, this is less of a problem when you can quickly verify the factuality of most responses. For most questions, Perplexity provides several citations, with a mixed bag in terms of authority. For example, for a typical academic question, Perplexity may cite and link you to:
A Wikipedia article
Encyclopedia Britannica
An accepted news outlet
An academic paper
A uUsingwebsite
Using Perpty for research can take you down some new paths you otherwise would have taken a long time to find. However, it’s still your responsibility to verify any information you see, because you are responsible for whatever you submit.
With responsible use, Perplexity offers great value for research utility per dollar spent. The free version is also very in-depth, with citations in many cases.
PDF Access
Perplexity can scan, understand, and summarize PDF documents. It can explain any topic in a way that helps beginners understand complex documentation. In-depth social science papers or volumes of code in a PDF file are broken down into their simplest representation.
PDFs are an important part of most online research processes. Most downloadable studies are hosted on websites in PDF format. So, this feature will more than likely be a game-changer in many research processes.
Transparency
With most AI chatbots, the citation is “trust me, but don’t.” The data in a chatbot’s answer can come from anything from a Wikipedia article to a leading research paper with thousands of citations from the author’s colleagues worldwide.
This doesn’t mean that Perplexity is a trusted source in itself. However, you don’t ever need to wonder why it answers the way it does. With different models, it may understand language and apply knowledge differently. But you can generally track the information it provides down to the source.
Perplexity Models
The default AI model on Perplexity is based on OpenAI’s GPT architecture. It offers the NLP, contextual awareness, and versatility of a ChatGPT model. The most noticeable difference is the addition of internet access and the citations feature. However, the free Perplexity model isn’t as up-to-date as GPT-4o or the other most recent models from other companies.
With a Perplexity Pro subscription, you can take advantage of Perplexity’s features while using other top-end AI chatbots.
Image credit: https://paperswithcode.com/ MMLU Benchmark
Claude 3.5
Claude 3.5 is one of the most up-to-date families of AI models by Anthropic. The Claude 3.5 Sonnet and Opus models are consistently among the highest-performing AI models on the benchmarks. On the MMLU benchmark, testing knowledge and problem-solving in 57 academic subjects, it is tied for third with GPT-4o.
With a Pro subscription, you can use the Perplexity search engine features with Claude’s performance and constitutional AI protections. This feature can make Perplexity even more effective as a research assistant. Constitutional AI from Anthropic provides stronger guardrails against inflammatory content and more principles leading to more useful content. All of this comes with the higher accuracy that Claude 3.5 demonstrates in text-based tasks, coding, and reasoning.
For complex but sensitive tasks of any kind, this addition adds a lot of utility to Perplexity’s features.
GPT-4o
Perplexity Pro users can also use GPT-4o. That means you get all the training benefits of GPT-4 models, including a performance that is even higher than Claude’s models.
Integrated with Perplexity, this latest model from OpenAI enables multitasking at even higher speeds.
Sonar
This Perplexity Pro option is integrated with SonarSource. In this case, you can use the LlaMa 3.1 70B open-source model with Perplexity’s other features.
According to Perplexity, this model was trained in-house to work well with the search engine.
Perplexity vs ChatGPT
We’ve already laid out the major differences between Perplexity and ChatGPT. The former is a search engine and research tool. The latter is purely an AI chatbot with additional features.
If you’re comparing these two options, you’re likely looking for an AI writing or research assistant. Perplexity offers more features and support for those purposes. However, ChatGPT continually stays ahead of the curve with its language models, and its API is easier to implement.
In this case, we can’t compare two equivalent chatbot models against each other. So, let’s go over the important factors that go into tools like these.
Coding
ChatGPT offers simple, streamlined code generation with text prompts. Overall, it is a simpler and more versatile tool for coding tasks.
Perplexity can also help you create code but has not been known to stand out for coding tasks. Perplexity’s search engine and other features weren’t meant to make it a better programmer. However, you can see examples of users reporting that Perplexity can be a great tool for checking code to ensure accuracy. Perplexity is also very good at helping you find other tools and resources for improving the effectiveness of your code.
Accuracy
Overall, Perplexity is better at accuracy due to its transparency and ease of fact-checking.
Neither ChatGPT nor Perplexity is an authoritative source for any kind of information. But that being said, both of them have advantages over their alternatives when it comes to accuracy.
Perplexity Pro users can apply the best AI models while also combing through Perplexity’s citations. Because Perplexity uses several models, we can’t pin a single overall accuracy statistic on it. However, it certainly is easier to verify and remove poor information.
GPT-4, and GPT-4o, in particular, are consistently among the leaders in the AI benchmarks. The MMLU, GSM8k, HumanEval, and MATH benchmark scores demonstrate that GPT-4 is a highly accurate AI chatbot.
The “expertise” that GPT-4 models demonstrate is multi-disciplinary. There are no major weak spots in ChatGPT’s knowledge. It does well in all text-based tasks, language, math, coding, and logic.
Perplexity vs ChatGPT: What To Be Aware Of
In some cases, Perplexity AI and ChatGPT can compete with subject matter experts. In some benchmarks that test up to the college level, GPT-4 scores above 90%. But that doesn’t mean that you can simply rely on the information that they produce.
Perplexity and ChatGPT both have disclaimers that indemnify them from the consequences of presenting the text they produce as fact. They can both make mistakes, and their content should not be copied directly and presented to employers or professors.
In addition, AI chatbots are trained on a huge mix of content available online. Sometimes, that training material is copyright-protected. News companies and academics have sued AI companies for using these materials without permission or compensation.
In some cases, it’s possible that AI-produced content includes text that can be construed as copyrighted. That’s why AI detectors like AI Detector also offer plagiarism detection.
ChatGPT, and the companies behind the models Perplexity uses, all face this issue to some extent. We can’t say for sure which option is better at avoiding plagiarism. But in the end, the fact remains that you take ownership over any content you submit, regardless of whether it was made with AI.
Last Word on Perplexity vs ChatGPT
Perplexity and ChatGPT are both great tools for research assistance. Perplexity shines more for its citations and access to the live web. The Perplexity search engine shows you exactly where it gets the information in its responses from and gives you a fast path to deeper research.
For general writing tasks, ChatGPT’s models are well-known for their language abilities and high performance when tested for knowledge and reasoning. ChatGPT is also a simpler and arguably more effective tool for coding.
In the end, both tools are very effective at a wide range of tasks. If you use them responsibly, either choice can be both entertaining and useful for serious tasks.
