*perhaps with help from my theory of propositions.
This post links to two astonishing chats with ChatGPT.
1. ChatGPT’s Philosophical Detection of Objectivist Epistemology
Model version: ChatGPT 4o
https://chatgpt.com/share/687c3047-2bcc-800b-bda8-28916181fde4
In this chat, the first two prompts and responses are of mild interest, but my third prompt was,
How do you relate a concept to the concrete instances that a concept subsumes?
From this simple prompt, ChatGPT 4o inferred that its response should rely on “the Objectivist theory of concept formation, developed by Ayn Rand, and elaborated by Dr. Harry Binswanger and Dr. Leonard Peikoff.” ChatGPT described Ayn Rand’s theory of concepts accurately and in depth, and contrasted her theory with other theories, including those that GPT relies on for its operation. ChatGPT then offered an account of what it would take to build a language model based on Objectivist epistemology. ChatGPT concluded, “This is not an LLM [large language model]. It’s closer to a cognitive robot with a rational, goal-directed epistemology — a genuinely artificial mind.”
This chat demonstrates that, despite resistance from academia, Ayn Rand’s Objectivist epistemology—supported by the work of Leonard Peikoff, Harry Binswanger, and others—has been weaved into the cultural mainstream.
2. Objectivist Language Model
Model Version: ChatGPT 5 Thinking
https://chatgpt.com/share/68aba04f-52a0-800b-b557-62e2173af339
When ChatGPT 5 came out last August, I started a new chat, asking, “Please explain how a language model based on Ayn Rand’s Objectivist theory of concepts would work.”
In this astounding chat, ChatGPT described “A Rand-style concept engine” and then a hybrid system combining the best features of the Rand-style engine and a standard LLM. The benefits of such a system range from more reliable and precise responses to a machine capable of making discoveries about the world and also about how the neural net is reaching its conclusions.
Although an LLM is prone to error, the careful reader can verify for himself the ideas expressed in this chat.
Some of the ideas in this chat are highly technical. One thing a reader can do is ask ChatGPT to explain an idea further. Just go to the bottom of the chat window and enter your own follow-up questions. Everyone with the link to the chat can continue the chat along his own line of questioning, independently of all the others with the link.
In related news, I have just filed a provisional patent on a proposed language system, as yet unbuilt, that draws on Ayn Rand’s theory of concepts but goes in a different direction from the system suggested by ChatGPT in the above chat. Stay tuned for updates.
4 thoughts on “Objectivist Epistemology* Might Revolutionize Language Models”
Very cool, ai must maintain a-is-a… nice. Building on their etymology of logic and reason, a common sense ai go figure.
Patent and copyright the first objectivist ai…? Capital hahahaaaa
Roll in your analytic equations for validating toward true and that could be quite a fun ai… Ron-in-a-bottle
Waiting excitedly to be an end user of your chat model. Congratulations
Ron, this is really intriguing. As far as I know, you are the only objectivist-oriented philosopher or scientist attempting to influence the evolving LLMs of AI. I have never thought of this until now, but it’s possible this is the most important line of inquiry for an objectivist philosopher, a project that could potentially have more positive influence than any other project. I am glad you are on the case, and look forward to hearing about your further progress.
I can see huge benefits from an AI that can perceive, identify and integrate sense-like inputs on the individual human hierarchical model and then further integrate parallel sense-like inputs in a manner that has never been possible previously. I envision an upload site that accepts audio-video inputs from millions of cell-phone an other digital cameras pointed at newsworthy and other events of interest and the AI generating moving 3D models of the events (more accurate the more POVs available) from which the AI can generate descriptions, sort by relevance to a user’s interests, and propose conclusions that can be verified by tracing back to the original inputs — GPS place-and-time-stamped. We already see the nascent version of this form of news report on television broadcasts that sometimes give the impression that every crime is video recorded, when simply those that happen to be recorded are those that will be covered in the video medium. Big Brother inevitably will be watching us but a long as we are watching back _en masse_we can retain freedom and security. In fact we are doomed if we’re _not_ watching back.
Comments are closed.