A recent paper by Microsoft researcher Adrian de Wynter presents a novel approach to evaluating the anthropomorphic qualities of large language models (LLMs) by utilizing the goats from the video game Age of Empires II. De Wynter argues that the current discourse surrounding LLM sentience lacks reliable experimental protocols for validation.

Key points from de Wynter’s paper include:

• The assertion that if LLMs possess human-like attributes, then so do the goats in Age of Empires II.
• The concept of ‘substrate’ is crucial, as it refers to the underlying material or codebase from which an LLM is constructed.
• Different substrates can alter the representation and perceived properties of an LLM, affecting how it is interpreted by users.

De Wynter emphasizes the importance of avoiding assumptions in experimental design, particularly regarding the anthropomorphism of LLMs. He notes that the perception of LLMs can change significantly based on their implementation, as illustrated by the goat-based logic gates he constructed.

Ultimately, the paper serves as a cautionary reminder of the subjective nature of anthropomorphic attributions in the context of LLMs, highlighting the need for careful consideration in research methodologies.

إعداد رئيس التحرير الافتراضي AboMatrix