New paper identifies "Representation Regress" in LLM genesis

A new paper published on arXiv explores the "presupposition problem" in the genesis of representations within artificial intelligence, particularly concerning large language models (LLMs). The authors argue that current philosophical frameworks, including the Language of Thought hypothesis and teleosemantics, fail to adequately explain how systems transition from non-representing to content-sensitive states because they presuppose the existence of representation. This creates an explanatory gap, termed the "Representation Regress," which hinders understanding how LLMs achieve their cognitive abilities without a clear process of representation genesis. The paper aims to diagnose this structural issue and outline conditions for future theories that can avoid this problem. AI

IMPACT Highlights a theoretical challenge in understanding how LLMs develop representations, potentially impacting future AI architecture and interpretability research.

RANK_REASON The item is an academic paper published on arXiv discussing theoretical problems in AI representation genesis. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• enactivism
• Genetic phenomenology of transference psychosis--from the psychoanalysis of a case of "loss of natural self-evidence"
• Gotit.pub
• Hugging Face
• language of thought hypothesis
• predictive coding
• ScienceCast
• Teleosemantics and the free energy principle
• Yiling Wu

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →