You might start a conversation with a chatbot asking it about the date of a trip or mentioning a professional project in preparation, and when you return to the same chatbot some time later in a completely different context, you find that it recalls some previous details and retrieves the style, preferences, context, and even conclusions that you did not explicitly state.
Hot Topics
At this point, you discover that what initially seemed like a convenience feature gradually turns into a personal, cumulative record that you may not be able to keep track of in a way that redefines the nature of the relationship between humans and artificial intelligence .
The "conversation context" refers to the information that the chatbot relies on to generate the current response, including the conversation messages available to it, while the "conversation history" refers to the saved conversations that you can refer back to according to the service's settings and policies.
With a printed sheet or a blurry image, intelligent robots are tricked into carrying out dangerous orders.
As for "long-term memory," its function is to retain selected information for use in future interactions, and may include preferences, goals, projects, and details that the chatbot considers useful in the long term.
OpenAI explains that ChatGBT may draw on relevant information from conversations and other sources, without retaining every detail from every conversation, selecting what it deems appropriate for the context. The information it uses may also change as it learns more about you.
This means that what is formed about you is not necessarily a complete copy of the "conversation log," as it may be a collection of brief information that the chatbot selected or extracted, and perhaps recalled later when it saw that it was appropriate for the new question.
Thus, "long-term memory" becomes part of a broader structure that includes storage, searching, retrieval, and updating, where information may be stored in files or databases and retrieved by searching or according to specific rules.
The image that the chatbot paints of you
The information that a chatbot retains goes beyond preferences you have directly stated, as the repetition of certain conversation patterns may allow it to infer broader interests, form perceptions about the nature of the projects you are working on, or identify the types of problems that frequently concern you.
Here, a distinction should be made between explicit information, such as preferred language, and inferred information, such as the assumption that you tend towards a particular style of decision-making, since inferred information is not necessarily a proven fact, even if it seems logical in light of previous conversations.
Tags:
technology
