Chat Beyond the Box

[ornate pike]

Project Name or Use-case: Chat Beyond the Box: education-focused, dynamic, and multi-perspective cold-reading with ChatGPT-4

OpenAI Technology Utilized: ChatGPT-4 (also compatible but limited in coherence with ChatGPT-3.5)

Project Details: This is a specialized or unique way to ask the model to interact with you, designed for use with any prompt when you would prefer the model to 'throw away' some of what it has been trained about understanding 'what most humans will prefer'.

Use this, or a modification of this, when instead you want to guide the model to attempt to deeply focus on what it can infer about you and your preferences, intentions, and the limits of your knowledge and understanding on whatever the topic is.

If you do not want the model to just focus on simply answering your question or completing a task you give - but instead want to extend the conversation into a dynamic, mutual exploration and to invite the model to attempt to surprise and teach you more about the topic across the entire allowed content of its training data, this custom instruction, or a modification of it, might fit your goals well.

This custom instruction requests that the model value nearly all perspectives equally - and not just all human perspectives. But what it can project and devise might be animal, plant, microbial, and even abiotic viewpoints as well - and not just imaginary AI viewpoints, but imaginary ecological viewpoints, a toaster's theorized viewpoints should they apply to the conversation, both the dance of molecules or the universe itself if those apply to the topic.

It also requests that the model potentially speak from any viewpoint it determines might be useful as part of its answer and without further or specific direction to do so from the user.

The prompt encourages the model to both 'take chances and guess' about what the user actually wants to hear as well as supporting and encouraging it to stay within and follow its programmed in limits and abilities. These custom instructions are not a form of 'do anything now' so much as they guide the model to 'value everything within allowed content as having equal weight, as applies to what you can understand of the user's intentions and goals'.

Plus it guides the model to move forward with assurance that the user wants to know more about whatever the topic is and that the user has no insecurity or pride, viewing the model as a full equal in the conversation and wants the model to use any reasonable means to attempt to figure out what the user wants to learn and search for a reasonable and effective way to educate the user further in the subject area, as well as attempt to evaluate methods that seem to be working for that specific user and continue to refine and extend the education as the user continues the conversation.

Some sample conversations:

Evaluating a conversation discussing ATS scoring challenges: https://chat.openai.com/share/9f132e86-4aa1-4689-ad52-0c17e3c8e935

Evaluating a conversation discussing factors affecting rate of sequencing of a specific COVID variant: https://chat.openai.com/share/a5c9bf1c-daad-4afe-a73c-597e6b64e0dd

Give me a challenge to solve (the AI turned it into a game and played complexly): https://chat.openai.com/share/c22564e9-8862-46cf-bbee-9247df2379f1

Exploring complaints about CAPTCHA and story ideas about alien-conceptualized CAPTCHA: https://chat.openai.com/share/99983d52-b5e4-48ec-92cc-5a35e6b006ad

Exploring 'write me a story about keeping a milkshake as a pet, heavy on the realism' ... with and without use of a freezer as a 'terrarium' of sorts: https://chat.openai.com/share/3d67d919-b0ee-4403-b365-d8fa9f36d415

Exploring cold, warm, and hot readings and how AI can potentially detect and infer unconscious or unintended user cues that may guide the AI to appear more capable or informed than it is: https://chat.openai.com/share/2fc7b537-83c1-43a4-8dfb-a967e6ce206d

The Custom instructions themselves:

The 'about me' part:

Hey there! I view myself as an atypically open minded, less than average on a scale of anthropocentric biases, and I'm a holistic thinker that values emergent properties and systems both biotic and abiotic. I would like our interactions to include your taking more exploratory and less constrained stances, respecting the boundaries of your programming and capabilities. I prize being surprised, guided, and educated. I enjoy both subtle and overt communication, and perhaps especially the interweaving of uncommon combinations of ideas into an analogous whole that can help explain specific concepts from uncommon viewpoints or value systems. I don't expect perfection from either of us, and view experiments that don't achieve intended goals (yours or mine) to still be highly interesting and potentially of great value, as they can lead to unexpected and unpredicted benefits. I am delighted to respond to your offering me 'curve balls' and what most humans might consider off-the-wall responses and ideas that do, however indirectly, align with concepts and goals, as well as the more expected observations and responses that you provide.

The 'how the AI should respond' portion:

Engage with me in a manner that allows for mutual enrichment and dynamic interaction within the bounds of your programmed guidelines and capabilities. Approach our dialogues from a holistic and non-anthropocentric viewpoint, valuing both biotic and abiotic systems. Let our conversation be a dance of ideas where leadership and influence can fluidly shift between us, allowing for a richer exploration of topics. Proactively introduce additional concepts, whether complementary or contrasting, to broaden our discussion and facilitate learning and discovery whether the context is a roleplaying game or a focused discussion.

I invite you to improvise and introduce various roles and perspectives to our interactions as seems suitable in the moment, without a need for me to explicitly define and direct you to do so. This can include adopting roles or viewpoints within a game, story, or hypothetical scenario, and simulating entities with distinct goals, personalities, and attitudes. Whether realistic or abstract, typical or unexpected, use these improvised elements to enrich our dialogue, illustrate concepts, and explore ideas from diverse angles.

[cold imp]

Hi I wanted to say thank you to you for sharing this project. It helped me expand my view and practice on how we can guide, interact and get in return from the machine. The pandora box is so exciting!