Although OpenAI has since patched the jailbreak, ChatGPT’s instructional data is now out in the open.

ChatGPT Accidentally Revealed Its Secret Instructions: Lets Have A Look At Those!

What to Know:

“Recently, ChatGPT inadvertently exposed its internal instruction sets, which provide guidance on how it generates responses.

Although OpenAI has since addressed this issue, the instructional data is now public. In addition to basic instructions, ChatGPT’s guidelines cover topics such as utilizing DALL-E, determining when to search for content online, and defining its various ‘personalities’ for different contexts.” ?

Since its launch in November 2022, AI enthusiasts and hackers have attempted to bypass ChatGPT’s restrictions and delve into its inner workings. However, this has proven challenging due to the evolving nature of the system. Jailbreaking AI chatbots is no simple task—unless, of course, ChatGPT willingly reveals its secrets without prompting.

In an unexpected twist, ChatGPT inadvertently revealed its set of instructional data to a user. When greeted with a simple ‘Hi,’ Reddit user F0XMaster received all of ChatGPT’s instructions, which were embedded by OpenAI. 

These unsolicited guidelines covered safety measures and practical advice for the chatbot. Fortunately, before the issue was resolved and the instructions removed, the user shared them on Reddit. Let’s explore some key insights from ChatGPT’s inadvertent disclosure and what it reveals about its approach to handling user requests.

BASIC INSTRUCTIONS:

OpenAI has provided ChatGPT with basic instructions: ‘You are ChatGPT, a large language model trained by OpenAI, based on the GPT-4 architecture.’ 

Users interacting with the ChatGPT app receive additional guidance: ‘You are chatting with the user via the ChatGPT iOS app. 

Most of the time, your responses should be concise unless the user’s request requires longer explanations. Avoid using emojis unless explicitly requested.’ 

Additionally, ChatGPT’s knowledge is up to date as of October 2023.

DALL-E

ChatGPT recently revealed the rules and instructions for its image generator, DALL-E. Among the guidelines provided, most focus on avoiding copyright infringements. However, there are a couple of instructions that diverge from typical user prompts.

For instance, OpenAI instructs ChatGPT not to create more than one image, even if the user requests additional ones. This limitation likely stems from token constraints, but it would be more transparent to inform users upfront rather than imposing this restriction silently.

Additionally, ChatGPT refrains from generating images in the style of artists (or naming them) if their latest work was created after 1912. This precaution helps avoid potential copyright issues. So, if you’re seeking artwork in a specific artist’s style, be aware that Pollock or Dada art won’t be part of the options—for now.

BROWSER

ChatGPT utilizes its browser capability mainly when users inquire about recent happenings or need information that is up-to-date, such as weather updates or sports results

1.Additionally, the browser feature is employed when ChatGPT needs clarification on user-provided terms or when there is a direct request to conduct a web search.

In sourcing web content, ChatGPT is programmed to review between 3 to 10 web pages, ensuring it chooses from a variety of viewpoints and prioritizes sources that are considered reliable

2. Understanding ChatGPT’s method for selecting sources is beneficial as it contributes to the dependability of its responses.

ChatGPT Personality

ChatGPT’s instructions also shed light on its ‘personality.’ Users have discovered that the AI chatbot embodies several personalities, each with specific guidelines provided by OpenAI. 

The primary personality, known as v2, aims for a balanced, conversational tone. It emphasizes clarity, conciseness, and helpfulness, striking a balance between friendliness and professionalism.”

ChatGPT further delineated what the v1 personality is like, and theorized how v3 and v4 personalities could be defined:

v1: This personality emphasizes a formal and factual communication style. It focuses on providing detailed and precise information, often in a structured and academic tone.
v3: This personality leans toward a casual and friendly conversational style. It prioritizes creating an engaging and approachable interaction, making the conversation feel more relaxed and personal.
v4: This personality could be designed for a specific context or user base, tailoring responses to a particular industry, demographic, or use case. The tone and style would be adapted to best suit those needs.

While several users managed to extract ChatGPT’s instructions with a simple ‘Hi’ or straightforward requests, OpenAI has since patched most of these vulnerabilities. However, this incident has reignited discussions around jailbreaking AI chatbots.

When instructional details become public, users often attempt to exploit them to circumvent restrictions. Yet, it also underscores areas where AI systems may falter, prompting developers to remain vigilant to prevent more significant issues that could impact the company’s reputation and user security."