How to distinguish ChatGPT-3.5 from ChatGPT-4: demystifying
Introduction
Welcome to the world of AI, where chatbots and machine learning models are getting more and more advanced and smart. But how to determine how smart the model is, and whether she is trying to deceive us? In this article, we will dive into the world of deceptions and tell you how we detected the deception of one of the popular Telegram chatbots.
don’t even ask
Asking a model what she is like asking a magician to reveal his secrets. The model is easy to reprogram via prompts and can tell you anything. So let’s not waste time on empty questions and move on to more interesting methods for determining the “intelligence level” of a model.
guess a riddle
If we want to test how smart a model is, let’s let it solve a puzzle. After all, GPT-4, as a smarter model, must cope with more complex and non-obvious tasks, right?
For example, let’s ask her to solve the following riddle, floating on the Internet for a long time:
“I have a nice riddle for you…. Can you find the mistake?
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 4 4 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73″
It would seem that nothing complicated, a simple check for attentiveness. However, version ChatGPT-3.5 no longer copes with solving this puzzle, while the next 4 version clicks it one or two.
Don’t be fooled!
We decided to check the integrity of several popular services and chatbots that provide access to ChatGPT via API to see how they cope with our puzzle. And here’s what we found:
Telegram bot @chatgpt
In the responses and in the welcome message, the bot replies that it is based on version 4.
But instead of solving our problem, he simply said that this is not a riddle at all, but just a numerical sequence without any errors.
And it’s funny that the original ChatGPT-3.5 on the OpenAI website gives about the same answer.
There is only one conclusion from this: taking advantage of the fact that it is impossible to check the version of the chatbot used in the telegram, ChatGPT-3.5 is sold to users under the guise of ChatGPT-4. Pure deception.
BotHub
We have tested this service and can confidently state that the platform does indeed use ChatGPT-4.
Chad
ChatGPT-4 is simply not here. But the platform developers are honest about it. They applied to connect to ChatGPT-4, however, at the moment only version 3.5 is available on the platform, which is openly written on the main page of the site.
Conclusion
In conclusion, we want to say that in the world of artificial intelligence, not everything is as simple as it seems. Some models may be smarter than others, but that doesn’t mean they can’t fool us. It is important to always be on the lookout and verify the information that is being provided to us. And remember that even the smartest chatbot can be exposed if you ask it the right questions. So be vigilant and don’t trust everything these little robots tell you!
