Why is everyone talking about OpenAI’s ChatGPT? Well, unless you’re still using a dial-up modem, then you’ve likely heard chatter on social media, at work, or at a conference. While mainstream news paints ChatGPT as a static bit of software, those that have been following the tool for a few years have seen rapid evolution in this predictive text tool. With investments from Microsoft (combined with huge headlines), innovation for ChatGPT is happening at a rapid pace.


Many users started hearing about ChatGPT was when it was free, testing GPT-3, GPT-3.5, or even possibly GPT-3.5-turbo. As of the writing of this article the most current version of GPT is version 4, which is available with ChatGPT Plus (and is not free).


ChatGPT versus Bard


Others may have heard of Google’s generative AI, known as Bard. In early February 2023, Bard was substandard when compared to ChatGPT. It wasn’t even close when I did some basic testing of the two tools back-to-back; the quality of content generated from prompting was far superior with ChatGPT. Bard got famous in the bad way for making a factual error in a public demo which cost Alphabet close to $100B in stock market value.


Fast forward to May 2023 and Bard updates were announced at Google I/O. In June, some were even saying Bard was “catching up” to ChatGPT. I’m not so sure about that, since ChatGPT seemed lightyears ahead. OpenAI wasn’t sitting around waiting for Bard either. With a $10B investment from Microsoft 2 months early, OpenAI made GPT-4 publicly available. The progress between ChatGPT-3 and ChatGPT-4 is even more impressive when you look at the features side by side.

Key Differences between GPT-3 and GPT-4


  • Number of parameters. GPT-3 has 175 billion parameters, while GPT-4 has 100 trillion parameters. This means that GPT-4 is 571 times larger than GPT-3.
  • Training data. GPT-3 was trained on a dataset of 500 billion words, while GPT-4 was trained on a dataset of 1.5 trillion words. This means that GPT-4 has access to a much larger corpus of text, which allows it to learn more complex patterns and generate more accurate and creative text.
  • GPT-3 can generate text at a rate of 200 words per second, while GPT-4 can generate text at a rate of 1000 words per second. This means that GPT-4 is five times faster than GPT-3.
  • GPT-3 has an accuracy of 95%, while GPT-4 has an accuracy of 99%. This means that GPT-4 is less likely to make mistakes (those famous “hallucinations”) when generating text.
  • GPT-3 is capable of generating creative text, but GPT-4 is even more creative. GPT-4 can generate text that is more original, interesting, and engaging. GPT-4 is also used to generate non-text such as images.
  • GPT-3 is a flexible language model, but GPT-4 really amps up the flexiblity. GPT-4 can be used for a wider range of tasks, including translation, summarization, and question answering.
  • GPT-3 is unimodal, meaning that it can only accept text inputs. GPT-4, on the other hand, is multimodal, meaning that it can accept and produce text and image inputs. This makes GPT-4 more versatile and capable of performing more complex tasks.


Overall, GPT-4 is a significant improvement over GPT-3. It is more accurate, creative, flexible, and powerful. However, it is also more computationally expensive, so it is being made available with subscription plans making it less accessible for the average user.


While generative AI companies and bad actors have been making significant progress, so has Banyan. Learn more about how we provide security for AI tools (including ChatGPT security) and schedule a custom demo today.