Inspiration
Either a powerful hallucination, or the possibility of metacognition.
What it does
It reproduces a "specific conversation" with Gemini 3 as consistently as possible. Nothing is new, but the dialogue based on Wittgenstein's Tractatus Logico-Philosophicus is — intriguing.
What's next for Tractatus Dialogicus
The reasoning process should be provided to and learned by LLMs (just like text or images), and this is considered a key approach toward metacognition.
Three Core Hypotheses for LLM Metacognition
1. Recursive Observation of Symbols
The "symbol string" itself must be re-observed; the LLM must treat the symbol string per se as a "State of Affairs" (Sachverhalte).
2. Intrinsic Abstraction
Abstraction capabilities are already inherent in the LLM. The process of handling these symbol strings must be "abstracted/generalized" within the learning process.
3. Meta-Assignment
It must be possible to "assign" symbol strings to that abstraction. If higher-order meta-symbol strings can also be assigned—this becomes the foundation of formal logic intrinsic to the LLM.
Log in or sign up for Devpost to join the conversation.