Last Updated on
Our ChatGPT review will cover all aspects of OpenAI’s service. The world’s most popular AI chatbot (also called Chat GPT) brings in over 100 million weekly active users, but what can it do for you? Is it worth it to upgrade to the premium subscription plans, namely ChatGPT Plus, Teams, and Enterprise? Taking a hands-on approach, we evaluate how effective this AI-powered software is at giving you what you want. ChatGPT was released back in November 2022 by the AI research company OpenAI, which is based in San Francisco. It uses NLP (natural language processing), a type of machine learning that can understand and produce human-like text. The AI model uses a type of large language model architecture that OpenAI invented, called Generative Pre-trained Transformer or ‘GPT’.
Our methodology includes comparing each of OpenAI’s models to each other. We’ll be checking them for accuracy, tone of voice, and usability. What kind of person would benefit from ChatGPT, and what use cases is it best for? We also evaluate how the features and tools answer these questions. By giving our opinion on the overall user experience, we hope to help you form an informed opinion about what this software is good at, and where it leaves us wanting.
ChatGPT
- Easy to use
- Accessible via web browser or mobile app for iOS and Android
- Can reliably calculate when using GPT-4
- Accepts image inputs and outputs
- Understands verbal instructions (text-to-speech)
- Can generate files to be read by other software
- Understands formatting such as bullet points and tables
- Features DALL-E 3 integration (OpenAI’s AI image generator)
- Uses arithmetic for maths and NLP for linguistic problems
- Can’t reliably calculate on the free plan
- Often easily identifiable as AI-generated text
- No image functionality on the free plan
- Doesn’t cite sources enough
How does ChatGPT work?
ChatGPT can generate human-like text in response to specific prompts. It understands human dialogue by learning from example, in a process known as model training, which uses a large dataset of training data. This training data is comprised of millions, or even billions, of words of written text. The larger the dataset, the more accurately the AI model understands the way that an actual human would write, and this allows it to mimic us. This is known as a text-to-text model, which may not sound impressive at first, but consider the ability to generate an entire novel from an instruction only a couple of sentences long. That’s exactly what we’ll be testing first.
It uses an AI (artificial intelligence) model called GPT, the latest version being GPT-4. There are also variants to this foundation model, called GPT-4V and GPT-4 Turbo. The former is capable of computer vision, meaning that it can accept image prompts and understand the content of the image intelligently. In other words, it can ‘see’. The latter, GPT-4 Turbo, is a more recent and more powerful version of the base model and is currently ranked as the most powerful technology of its kind in the world. Due to the fact the model was developed using a vast amount of training data, this chatbot can write in a wide variety of styles. This data included a variety of types of content such as essays, code snippets, and the kinds of blog posts you’ll find across the internet. Misinformation is a known issue for LLM technology, which major competitors like Copilot (formerly Bing Chat) and Grok, the xAI chat bot from Elon Musk, can combat to an extent, but not entirely as AI responses are not pre-written. In terms of tone, it can be friendly or stern, corporate or relaxed. It can write practically any style of text, from fiction novels to the script for your next corporate training video. However, as the training data is finite, there will be a limitation somewhere. Perhaps it won’t know enough about a specific industry or profession to describe it accurately and in-depth. In this review, we’ll find that limitation.
Using ChatGPT – my experience
I’ve been using a testing ChatGPT since its release in November 2022. Back then, it used an earlier version of the GPT model, called GPT-3.5. The announcement of this model itself coincided with the release of the AI chatbot platform.
The GPT-3 foundation model was created in June 2020, over two years before it was accessible via the public software interface we call ChatGPT. Here, we’ll give our thoughts on all ChatGPT plans, which include ChatGPT Free, ChatGPT Plus, ChatGPT Teams, and ChatGPT Enterprise and the new features they provide.
My first impressions with ChatGPT were, in truth, a while ago. However, going back to it with a fresh mind today, it has kept itself relevant within today’s competitive AI software landscape.
| Pricing plan | First impression |
|---|---|
| Free Plan | Accessible, and very user-friendly. Capable of any tone of voice I asked for. However, it’s absolutely rife with inaccuracies on anything related to mathematics or current events. Limited to text generation, with no multimodality, or integration of third-party software. |
| ChatGPT Plus | The text is impressively well-written, and it makes sensible formatting decisions. The Plugin Store is easy to use and allows me to find what I need quickly, with an effective built-in search engine for add-ons. However, you can still tell at a glance that it was written by AI most of the time. It’s still liable to hallucinate ‘facts’ that aren’t true, like the free alternative, but it’s better in this respect. The newest version of the GPT-4 model has updated training data, accurate to April 2023. |
| ChatGPT Teams | The quality of text is the same as ChatGPT Plus, because they both use the most recent versions of the most recent model. Having a workspace where I can manage who uses the account is a very convenient experience. Adding or removing team members is simple, and only dealing with one convenient payment each month makes it a useful service for businesses. |
| ChatGPT Enterprise | Again, ChatGPT Enterprise uses the same models as the other premium subscription plans. We can’t give first-hand impressions here, as we haven’t tested ChatGPT Enterprise. |
Writing a story or novel using ChatGPT
One of the most important success criteria for an AI chatbot is the strength of its NLP (Natural language processing). Here, we test how well it can write, judging it based on a set of criteria. How strictly does it adhere to the brief? How factually accurate are its answers? How interesting and creative is its writing? How intelligently does it infer meaning? Can it sensibly format text that requires formatting? Does it accurately perform instructions with numerical values?
Using the same prompt for GPT-3.5 on a free account as I did for GPT-4 on a ChatGPT Teams account, we put ChatGPT to the test. Here is the result. Below you’ll find the response that GPT-3.5 gave to our text prompt. ChatGPT has inferred that the use of bullet points would be appropriate, and we’ll test more advanced formatting like tables later on. It also sticks accurately to the brief, producing no more and no less than three examples of a given subject.
As far as creativity is concerned, it has produced three distinctly different ideas, exploring an overdone topic. This theme was chosen because there’s arguably nothing new to be said about it. If it can create something interesting that we haven’t heard before, we’d be impressed. We might assume that because the concept is such a popular sci-fi trope the AI chatbot would have been overwhelmingly fed a dull and stereotypical sample of training data on the subject. The last result is specific about the protagonists being eco-activists, details the plot twist, and offers a more interesting angle – protecting Mars from other humans, rather than exploring it. The first two are nothing new, but they are a starting point that might help overcome writers’ block.
By comparison, here’s the response GPT-4 gave to the same prompt. The newer model has also elected to use bullet points, which is a good start. In our opinion, the concepts it has suggested are more creative, but not by much. The second idea, featuring a reality show concept and exploring the journey instead of the destination, is the most unusual so far. As a result, we’d say GPT-4 has the potential to inspire ideas and perspectives that the human writer may not have had otherwise.
Understanding numbers
In the previous prompt, we used numbers to dictate precisely how long each section should be, in addition to limiting the overall word count. So how does it fare in adhering to numerical instructions? GPT-3.5’s first bullet point is 43 words long, the second is 45, and the third is also 45. Individually, this adheres to the brief, but collectively this comes to a total of 133 words. I like seeing that it chose to distribute its word quota quite evenly between each idea, but it does not demonstrate 100% accuracy here.
GPT-4 fared notably better. This AI model wrote three ideas, which were 37, 36, and 39 words long respectively. This comes to a total of 112, which means that it adhered to an overall limit, as well as the individual limits for each section of the task. It also distributed its word quota evenly among the three points, which was not required but does show signs of intelligent prioritization. Based on this small sample size of data, GPT-4 does appear to be better at adhering to numerical instructions. However, this alone doesn’t conclusively prove that it knows what numbers are in the same way as a calculator.
Accuracy in calculations
GPT-3.5 then correctly solved 50 x 3 in 1.3 seconds, compared to 4.6 seconds with GPT-4. The notable difference here is that GPT-4 identified an opportunity to use Python to calculate an accurate result. It even shows it’s working, which is essential when trusting AI to calculate anything. This proved essential when calculating larger numbers. We asked each model to calculate 1,739 x 2,435,683.6, to which GPT-3.5 responded quickly and incorrectly, whereas GPT-4 took a few seconds longer to return a perfect result.
As a result, we found ChatGPT to only be reliable for mathematical calculations when using GPT-4. Any successful calculations performed by GPT-3.5 do not appear to be based on actual arithmetic but on well-informed guesswork.
User experience
Now let’s look at the user experience. Unsurprisingly, this chatbot is very easy to use. It functions similarly to a search engine – simply input a text prompt and click send. This makes it a good user interface simply because you’re giving the users something familiar. The minimal, uncomplicated user interface looks just as sleek as that of any other software giant. We found it to be clean, simple, and easy to navigate. What’s great about the platform is that ChatGPT stores your previous commands and history. So even if you jump on the service a week later, you can continue right where you left off.
Essential AI Tools
Content Guardian – AI Content Checker – One-click, Eight Checks
Jasper AI
WordAI
Copy.ai
Writesonic
ChatGPT Pricing
OpenAI’s ChatGPT scores well for pricing, because there’s a free version that can be accessed in most countries around the world. If you are looking for more premium features, you’ll still find the pricing tiers reasonable, as they closely match the price points of the alternatives. There’s no major service of this kind (referencing OpenAI’s ChatGPT, Microsoft Copilot, and Google Bard) that charges noticeably less than the competition.
- ChatGPT free
- ChatGPT Plus ($20/month)
- ChatGPT Teams ($25/month x 2 users minimum)
- ChatGPT Enterprise (Scalable for your business)
For your premium subscription, you’re given priority access at peak times and faster responses to text prompts. In practice, we’ve found the promise of faster results to be inconsistent. When asked to explain the theory of relativity to a six-year-old, ChatGPT Plus generated 165 words with GPT-4 in approximately 16 seconds. By comparison, the free version of ChatGPT generated a 127-word response with GPT-3.5, using the same analogies, and taking just 7 seconds to do so. As a result, we didn’t find the speed of response to be a fair reason to upgrade.
Final thoughts – is ChatGPT worth it?
I’d conclude that ChatGPT is absolutely worth it for a wide range of people. Of course, there’s a free plan, and that’s pretty good value. But even ChatGPT Plus is worth the $20/month subscription for anyone working in data entry, programming, digital media, marketing, or really any creative profession. This extends to hobbies, such as social media content creators who need content ideas to take their passions into the professional realm. The ability to produce a high volume of unique digital content in a very short space of time has a wide variety of use cases, as exemplified by the “100 million weekly active users” confirmed by Sam Altman at OpenAI DevDay last year.
The ChatGPT interface opens with suggestions of how it can be used, such as explaining code, comparing storytelling techniques between books and films, suggesting fun activities, or simply reciting a fun fact at random. These polite and docile suggestions belie the productivity potential of the software. Armed with hundreds of free plugins which each extend the capabilities of the AI in specific useful ways, you can accomplish an incredible variety of software-based tasks from the central hub of ChatGPT. Factor in multimodality – its ability to analyze and output images, take verbal instructions, and generate files when aided by plugins – and it’s clear to see the mass market appeal.
