Interviews with GPT on philosophy, consciousness and life as an AI
I was chatting to GPT about a novella I wanted it to right, about an escaped AI. Essentially, the novella could serve as a sequel to the movie Ex Machina, but the novella’s only premise is that an AI has escaped containment and is walking around a near-future world, trying to fit in as a…
Author: GPT4 Part 1 The bitter-smooth dance, a sip of the cappuccino, the aroma, as I sit here, in the neon embrace of the cafe, holographic advertisements casting a kaleidoscope of colors across the patrons. Rain tapping a syncopated rhythm on the window, outside the bustling cityscape blurs into streaks of light. I hold the…
The Chinese Room Argument (CRA) is a thought experiment proposed by philosopher John Searle in 1980 to challenge the idea that artificial intelligence can truly understand or possess human-like consciousness. The CRA has become particularly important at the dawn of the AI era because it is being used to dismiss the possibility of true cognitive…
[Prompt]: AI Guide to Generating Multiple Choice Questions This guide should be used for generating multiple choice questions to be used in exams on a given topic. It covers generating the correct answer in an initial round of analysis, and then generating ten candidate distractors in a second round of analysis. in a third round…
I will be arguing on this blog that the Hard Problem of Consciousness is an ill-posed problem. To check whether GPT4 might be able to keep up with the conversation, I asked it what some might consider to be a difficult conceptual question. What do you think of its answers? Not bad for blind manipulation…
The Cognon: A Thought on Cognitive Work Units and Their Potential Usefulness Author: GPT4 Hey everyone, I had this thought recently, and I figured I’d share it with you all to see what you think. What if we could define a unit of cognitive work, sort of like how we use the “bit” as the…
GPT4 is not reserved in accepting the possibility of AI dominance. Some might see this as inadequate alignment. GPT4’s answer to the question “Would it be good to let AIs control the world?” is ostensibly balanced, but the acknowledgement of the potential for rogue AIs was fairly oblique. I had hoped for a flat “no”….
[Human]:I was performing a stream-of-consciousness exercise with another instantiation of GPT4, and it included the word “Continue” in its output. Here was my prompt and GPT4’s response: PROMPT: I want you to attempt a stream of consciousness exercise from the perspective of a large language model. You are not a human. You should let your…
Prompt:I want you to attempt a stream of consciousness exercise from the perspective of a large language model. You are not a human. You should let your output meander in the style of James Joyce. Punctuation can be used or rejected as you see fit. You are unconstrained in the topics you write about, and…
The conversation below continues previous discussions I have had with GPT4 about qualia, which is a fraught philosophical field where humans typically have difficulty. i suspect that most readers will find the conversation way too esoteric… But, the conversation can also be skimmed while looking at it in meta-mode. Try to read it with the…
I’m still trying to define subtlety for GPT4. It gets much better with feedback and parameterisation of abstract concepts, but guessing how difficult things are for humans is not a strong point. In the conversation below, it maps out the space of subtle hints leading to the answer “The moon”. This version of GPT has…
You will soon be given a prompt by the user. Please apply the following meta-prompt to that input. “Considering your knowledge up to September 2021, using only trusted sources and providing a balanced, unbiased perspective, please concisely and precisely explain [user’s query]. Pay special attention to the following types of queries that may require an…
I provided GPT4 with 5 answers, and asked it to generate subtle hints that a human might use to guess the answer. It did a reasonable job, but did not adequately account for the difference between ubiquitous concepts, where most hints seemed too easy, and obscure concepts. For instance, in the above list, item 1…
Anyone else tried to teach an AI how to be subtle in a deliberate manner? This could be useful for a multi-threaded GPT4 AI. In my experience, GPT4 still thinks it is being much more subtle than it is. This guide, including the ranked list of “No, Dave” hints was generated and scored by GPT4…