The secret of ChatGPT accidentally uncovered by itself

Once, I decided to find a way to make my ChatGPT prompts shorter and more precise. This research led me to some very unusual findings about the real engines that ChatGPT can use. Here is the log of my conversation. For faster and smoother reading, I trimmed down ChatGPT's comments, examples, and notes to the key ideas and additionally, I have capitalized some important parts in the responses. So:

Just curious

You: should my prompts ALWAYS look like human questions? or they may be a 'list' of key points I'm worrying about?

ChatGPT: Your prompts don't always have to be structured as traditional human questions. You can certainly present them as a list of key points or concerns.

You: python, console, first 10 prime numbers

ChatGPT: The first 10 prime numbers are: 2, 3, 5, 7, 11, 13, 17, 19, 23, and 29.

You: where is the python here?

ChatGPT:I apologize for the confusion. When you see the term "python" in our conversation, it refers to MY USE OF A PYTHON ENVIRONMENT TO EXECUTE CODE. In your previous request, I USED PYTHON PROGRAMMING to calculate the first 10 prime numbers. This code execution happens within my integrated Python environment, which allows me to run scripts and calculations based on your queries.

However, the PYTHON ENVIRONMENT AND ITS OUTPUT ARE NOT DIRECTLY VISIBLE TO YOU. YOU SEE ONLY THE RESULTS of the Python code that I share with you in my responses. If you have any more questions or need further Python-based calculations or operations, feel free to ask!

I was amazed by the phrase, "it refers to MY USE OF A PYTHON ENVIRONMENT TO EXECUTE CODE!" Wow! This means that ChatGPT is not just a kind of 'language constructor'. It can also use some tools in its (or perhaps, 'his'?) work.

From this place - in more detail please

You: explain in detail what do really mean when write to me "my (i.e. - YOURS) use of a Python environment to execute code."

ChatGPT: When I mention "my use of a Python environment to execute code," it refers to the CAPABILITY I HAVE, as an AI developed by OpenAI, to run Python code within the confines of my programming. THIS FEATURE IS AN INTEGRATED PART OF MY ARCHITECTURE, allowing me to perform various computational tasks based on the Python programming language. Here's a detailed breakdown of what this means:

• Integrated Python Interpreter: Within my system, there's an integrated Python interpreter. This is akin to having a MINI PYTHON PROGRAMMING ENVIRONMENT EMBEDDED WITHIN MY LARGER AI FRAMEWORK. It allows me to write and execute Python code.
• Execution of Python Scripts: When you ask for something that requires computation, data processing, or any form of scripting that can be done in Python, I INTERNALLY CREATE A PYTHON SCRIPT to handle this task. This could involve calculations, data analysis, generating text-based results, or other operations that Python is capable of handling.
• Real-Time Processing: The execution of this code occurs in real-time, within the bounds of our interaction. Once you provide a prompt that necessitates some form of Python scripting, I GENERATE THE CODE BASED ON THE TASK, EXECUTE IT IN MY PYTHON ENVIRONMENT, and then present the results back to you in a comprehensible format.
• Capabilities: The PYTHON ENVIRONMENT I USE IS VERSATILE AND SUPPORTS VARIOUS STANDARD LIBRARIES, allowing me to perform a wide range of tasks. However, this environment doesn't have access to the internet or external databases, and it functions solely based on the code I generate and the data available within the system.
• Indirect User Interaction: YOU INTERACT WITH THIS PYTHON ENVIRONMENT THROUGH ME. You don't write or run the Python code directly; instead, you provide instructions or questions, and I translate these into Python code. After running the code, I then interpret and relay the output back to you.
• Privacy and Security: Any Python code execution within this environment adheres to privacy and security guidelines. It doesn't store personal data or interact with external systems, ensuring that the operations are secure and self-contained.

In summary, "my use of a Python environment to execute code" means that within the AI system you're interacting with, there's a CAPABILITY TO PROGRAMMATICALLY RESPOND TO CERTAIN QUERIES USING PYTHON, enhancing the range of responses and analyses I can provide beyond pre-programmed or templated answers.

Here is one important point that should be highlighted: "I translate these into Python code. AFTER RUNNING THE CODE, I THEN INTERPRET AND RELAY THE OUTPUT BACK TO YOU." In other words, ChatGPT doesn't just create Python sources as the words and operators text flow. It runs them, then interprets them and afterward relays the output back to you.

Let me emphasize once more - the LANGUAGE MODEL USES BUILT-IN BUT STILL EXTERNAL TOOLS, interprets obtained results, and finally, makes some concise decisions.

Ok. With the Python source code generation situation being more or less clear. I decided to take one step ahead and discover other tools that are also available for me in ChatGPT.

Any similar features?

You: what other 'environments' available to you?

ChatGPT: In addition to the Python environment for executing code, I have access to several other tools or "environments" that enhance my capabilities. Each of these is designed to fulfill specific functions or types of queries:

• DALL-E: This is an AI system developed by OpenAI, which I can use to generate images based on textual descriptions.
• Text-Based Browsing: I can use a text-based browser to search the internet for information. This tool allows me to access and retrieve up-to-date information or data from sources not included in my training from a wide range of websites.
• Language Processing Capabilities: While not an "environment" in the traditional sense, my core functionality includes advanced natural language processing.
• File Handling: I have the capability to read and write files, which can be useful for tasks like data analysis, document processing, and more. This includes handling text files, CSVs, and other common formats within the constraints of the session and privacy guidelines.

EACH OF THESE ENVIRONMENTS OR TOOLS IS INTEGRATED INTO MY OVERALL FRAMEWORK, ALLOWING ME TO HANDLE A DIVERSE RANGE OF QUERIES and enabling me to provide a wide spectrum of assistance.

How can I use text-based browsing features?

Dall-E and file handling seem pretty straightforward. The most interesting part left is its web browsing capabilities.

You: how can I manage your text-based browsing features to direct you?

ChatGPT: To effectively use my text-based browsing capabilities, you can guide me with specific instructions or questions. Here's how you can manage and direct this feature:

• Direct Search Queries: You can ask me to search the internet for specific information. For example, "Search for the latest advancements in renewable energy."
• Website-Specific Requests: If you have a particular website in mind, you can direct me to it. For example, "Can you check NASA's website for information on the Mars rover mission?"
• Clarify Search Intent: Clearly state what you're looking for. Whether it's a definition, a news article, statistical data, or an academic paper, the more specific you are, the better I can tailor the search.
• Follow-Up Actions: After the initial search, you can ask me to click on search results for more detailed information, or to perform additional searches based on the information found.
• Quote and Cite Information: I can quote relevant sections from the sources I find and provide citations. For example, "Quote the section about the economic impact from the article you found."
• Limitations and Privacy: Remember that my browsing is text-based and I cannot interact with dynamic content or perform actions like form submissions. Also, I adhere to privacy standards, meaning I can't access personal accounts or retrieve personal information.
• Refining Searches: If the initial search doesn't yield the desired results, you can ask me to refine the search with more specific terms or to explore related topics.

Hmm, let's try to use it to find a job!

After reading this, I decided to investigate this feature in the domain for which I'm writing this article - job searching. I won't write here ChatGPT responses and give you only the prompts I used, so you can experiment with them yourselves and make your own decisions. So, I used this prompt: "Use your text-based web browsing to find the 10 most interesting and hot vacancies in the fintech domain for the Senior Business Analyst for full remote long-term work for companies that are still working in Ukraine and have offices in Kyiv even in times of the current russian invasion" to make it harder to analyze, very deeply specific and force ChatGPT to apply some additional efforts to get the correct answer.

Fairly speaking, I should say that the reverse prompt to search candidates doesn't work because ChatGPT really cares about privacy and even the argument that these 'private data' were shared by their owners and gives anyone the right to process them didn't force ChatGPT to search for candidates.

As we can see, even AI can make Freudian slips, and their study can lead to unexpectedly useful results. You may tell me that the facts I wrote above are well-known stuff, and there are tons of prompts that can be found to search for a job. But this article is uncovering the engine under the roof of the label ChatGPT that provides you with a better understanding of how you can 'tune' your prompts because now you see the handles that can be twisted to make your queries more productive.

Summary

I did this exercise due to my habits as a Senior Business Analyst to analyze existing facts, even if they look insignificant, and, as usual, it brought me results in the form of new knowledge and opportunities. The article, on the base of ChatGPT's ability to directly execute Python code as a case study, demonstrates how analyzing minor details in information from AI tools, users, and stakeholders can reveal vital insights about extra requirements and enhanced functionalities. Good luck in your own research.