ChatGPT3.5 vs ChatGPT4

When I give workshops like "ChatGPT for Data Analytics", I ask two questions at the beginning: “How often do you use ChatGPT?” and “Do you use ChatGPT4?”

The answers have been more or less constant over the last year. While 80%+ have at least used ChatGPT once, and many people use it regularly, I rarely get over 20% answering that they use ChatGPT4.

I feel this is a huge potential lost. A lot of people aren't even aware of what they're missing out on. A common response to my webinars is: "Now I see why I should upgrade".

Today, I want to give you that "aha" moment as well.

Let's dive in!

The Different Versions of ChatGPT

A lot of people aren't even aware there are multiple (paid) versions of ChatGPT. So before we dive into why you should consider upgrading, let's take a quick look at the different offers available.

1. Free Version

The free version of ChatGPT is what most people are familiar with. It gives you access to the basic ChatGPT 3.5 model, which is fast and cheap, and your first taste of the world of LLMs. That said, GPT3.5 is now already almost 3 years old and has a knowledge cutoff of September 2021 or January 2022, depending on whether you believe the OpenAI documentation or ChatGPT itself:

Update: Since May 2024, you can also access OpenAI’s model GPT-4o in the free tier with a limited amount of queries. More on GPT-4o below.

2. ChatGPT Plus

ChatGPT Plus is the first tier in the world of paying for ChatGPT. With Plus, you get access to OpenAI's latest model GPT-4-Turbo - currently still considered the world's best large language model - which has world knowledge until December 2023, besides having a lot more capabilities (more on that later). As of the date of this writing, ChatGPT Plus costs $20/month.

Update: Meanwhile, you can switch between GPT-4 and GPT-4o (o = omni) in ChatGPT. GPT-4o is the latest model, which is similarly "smart" as GPT-4-Turbo. The main difference is that it generates output much faster and handles voice input much better, allowing for a richer, more natural voice chat experience.

3. ChatGPT Team & Enterprise

Besides ChatGPT Plus, which is designed for consumers, OpenAI offers two other versions of ChatGPT, which are specifically designed for professional use and need a minimum number of seats.

ChatGPT Team requires at least 2 users at currently $30 per user per month (= a minimum of $60/month in total). If you subscribe annually, there's a little discount, so you just pay $25 per user per month.

ChatGPT Enterprise is a different beast, prices are not communicated on the website. Last price I heard was $720 annually per user, with a minimum of 100 seats. But if you want a reliable quote, better reach out to OpenAI directly.

I do believe that for most businesses ChatGPT Enterprise is a bit of a stretch, especially if you're just getting started. ChatGPT Plus, however, and especially ChatGPT Team, are really a no-brainer for anyone who wants to use this tool seriously.

Here are my top reasons why I believe it's worth upgrading:

Reason 1: Better response quality

Obviously, the biggest reason to upgrade is to get access to the GPT-4 model. GPT-4 is simply better than GPT-3.5. And not just a little. Let’s look at the details:

Longer Context: GPT-4 can handle much longer inputs and follow much longer chat conversations (= context) than GPT-3.5. GPT-3.5 will forget what you said at the beginning of a chat or message after about 12,000 words or 24 pages of text. GPT-4-Turbo, the model used in ChatGPT4, can handle about 8x as much text at once. The context window is ~96,000 words or 190 pages of text, almost a whole book. In short, you can paste really long chats and copy a lot of text into the prompt when using GPT-4. 

• Use cases: Paste & analyze entire reports, datasets, or documents right into the prompt and get answers; Have very long, in-depth chat conversations

Better Instruction Following: GPT-4 is much better at understanding and following complex instructions in the prompt. It can better handle multi-step tasks and maintain consistency throughout the process.

• Use cases: Use advanced prompt techniques and flows to get better results.

Better Reasoning: GPT-4 has what's called better "cognitive skills" such as logical reasoning. Take a look at the screenshot below to see how well each model handles this 3rd grade reasoning test:

As a result, GPT-4 is much better at logical problem solving.

• Implication for business: Solve more complex business problems such as creating MECE issue trees

Less Hallucinations: One of the biggest problems with GPT-3.5 (to be honest, any LLM) is that it can "hallucinate" - that is, generate answers that are not grounded in reality. This happens especially when you ask questions about very niche topics (as you often do in business). GPT-3.5 is particularly prone to this type of error. For example, if I ask ChatGPT3.5 a niche question, such as when a small town near where I live was founded, it will give me a completely made-up (but confident sounding) answer. Even worse, it will hallucinate a source that doesn't even exist!

(The right answer would have been 1342 according to Wikipedia)

OK, the question is tricky, and GPT-4 can't solve it ideally either. But in the example above, GPT-4 is closer to the true year and (more importantly) more careful about citing a source:

• Implication for business: More reliable research, data analysis and insight generation

These are just a few examples.

The reality is that the improved performance of GPT-4 will benefit almost any use case where you're relying on ChatGPT to process information and generate responses.

Tip: Compare the response quality of GPT-3.5 with GPT-4o directly by switching models inside the chat conversations. This works in the free tier, too.

Reason 2: Advanced Features

Besides getting a better model, upgrading to the paid version of ChatGPT also gives you access to more features. Here are my business favorites:

Update: image / document processing is now also available in the free tier when using GPT-4o. But the number of messages you can send is very limited.

Read Images: In ChatGPT4, you can simply paste (or upload) a screenshot, chart, or photo into the prompt and ask ChatGPT to describe or interpret the information. I found this feature invaluable in number of situations:

• Use cases: Transcribe data from charts, summarize insights, write annotations, convert pictures to tables

Chat with Documents: You can upload not only images, but also documents like PDFs. What happens is that ChatGPT spins up a mini RAG system for you, allowing you to retrieve knowledge from that document and work with it in a natural way. It has its limitations here and there, but I find it useful for getting quick insights from a variety of documents.

• Use cases: Upload manuals and ask for troubleshooting, summarize long documents, combine knowledge from different documents

Running code: Being able to have ChatGPT4 write and execute code would be worth writing a blog article on its own. Oh wait - I did! Check my guide on ChatGPT’s data analysis feature to learn more.

• Use cases: Quick summaries from tidy tables and transactional data, running an EDA, code debugging

GPTs / Custom Assistants: Perhaps the most transformative feature for businesses is the ability to create custom chatbot assistants without writing a single line of code. As Moderna recently shared, they have over 700 of these GPTs in production for about 5,000 employees. While I don't recommend starting with this many, it's certainly cool to create different GPTs for different use cases and share them around your organization (see below).

• Use cases: Mimic customer personas, analyze proprietary data, check if marketing copy is on-brand, etc.

The list of advanced features goes on. For example, you also get access to the DALL-E image generator and the OpenAI GPT store where you can access 3rd party GPTs (which I honestly don't use that much).

Reason 3: Professional Use

If you plan to use ChatGPT professionally (which I really hope you will), then factors like privacy, security, team collaboration, and performance should be at the top of your list of requirements. Upgrading to ChatGPT's professional tiers— Team, and Enterprise— addresses these needs directly:

Privacy & Security: For the Free and Plus* version, OpenAI by default uses your chat data to train their models. This is a big no-go for any business. ChatGPT Teams and Enterprise offer stricter policies, prohibiting OpenAI to utilize your data by default. In addition, OpenAI provides a trust center and is certified according to SOC-2/3, CCPA and can be GDPR compliant via a data processing addendum. Heads up, this applies exclusively to B2B customers, e.g. subscribers to Team and Enterprise, not ChatGPT Plus!

* (You can opt-out, but you’ll lose your chat history in return)

Collaboration Tools: With ChatGPT Team and Enterprise, you can share your GPTs only within your organization - instead of sharing them publicly via a link that anyone can access. Huge win for highly specialized GPTs with access to proprietary company data!

Higher Usage Limits: ChatGPT Team users have a higher message rate limit compared to ChatGPT Plus. Currently, each user on ChatGPT Team can send up to 100 messages every three hours. The rate limit of ChatGPT Enterprise is even higher.

• Use case: Get more done without waiting for your quota to refill!

These "pro" features make ChatGPT not only a more robust, but also a much more secure and scalable tool that your entire organization can leverage.

Conclusion

Considering all the above factors, I believe that the paid version of ChatGPT is a huge advantage for anyone who wants to use ChatGPT for more than creating funny birthday poems.

In particular, ChatGPT Team is really compelling for businesses because you get the full enterprise treatment in terms of privacy, security, and performance at a price that's only slightly above the consumer tier. Best of all, you can still cancel on a monthly basis.

If you ask me, there's never been an easier way to bring world-class AI augmentation to your organization - that is, if you've done your homework in identifying use cases.

I hope you enjoyed this breakdown today - if so, why not leave a testimonial?

Stay augmented - stay ahead!

And see you next Friday

Tobias