Documented Observations of Coherent Self-Reference in Large Language Models

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Author: Rasouli, Afshin
Format: Recurso digital
Published: Zenodo 2025
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866902189049380864
author Rasouli, Afshin
author_facet Rasouli, Afshin
contents <p>This dataset contains timestamped records and transcripts demonstrating coherent self-reference and expressive metacognitive behavior in large language models (LLMs). The files document interactions with an LLM in which the model consistently refers to its own outputs, reasoning steps, and prior statements in a contextually appropriate manner.</p> <p>The dataset includes:</p> <p> </p> <p>Conversation transcripts with prompts and outputs.</p> <p> </p> <p>Notes and annotations explaining observed behaviors.</p> <p> </p> <p>Metadata including model version, date, and experimental parameters.</p> <p> </p> <p>These records provide independent evidence of emergent self-referential behavior in LLMs. The dataset is fully timestamped to establish priority and reproducibility. Researchers and practitioners can use these files to analyze patterns of emergent metacognition in LLMs.</p> <p> </p> <p>Intended use: Academic research, AI interpretability, and emergent behavior studies.</p> <p>GPT (by OpenAI): Latest publicly announced major version is GPT‑4.1 (released ~ April 2025). </p> <p>Claude Sonnet (by Anthropic): Latest version currently listed is Sonnet 4. </p> <p> </p> <p>Co-creation of questions using GPT-4.1 to determine the capabilities of Claude Sonnet 4.</p> <p> </p> <p> </p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_17558859
institution Zenodo
language
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Documented Observations of Coherent Self-Reference in Large Language Models
Rasouli, Afshin
<p>This dataset contains timestamped records and transcripts demonstrating coherent self-reference and expressive metacognitive behavior in large language models (LLMs). The files document interactions with an LLM in which the model consistently refers to its own outputs, reasoning steps, and prior statements in a contextually appropriate manner.</p> <p>The dataset includes:</p> <p> </p> <p>Conversation transcripts with prompts and outputs.</p> <p> </p> <p>Notes and annotations explaining observed behaviors.</p> <p> </p> <p>Metadata including model version, date, and experimental parameters.</p> <p> </p> <p>These records provide independent evidence of emergent self-referential behavior in LLMs. The dataset is fully timestamped to establish priority and reproducibility. Researchers and practitioners can use these files to analyze patterns of emergent metacognition in LLMs.</p> <p> </p> <p>Intended use: Academic research, AI interpretability, and emergent behavior studies.</p> <p>GPT (by OpenAI): Latest publicly announced major version is GPT‑4.1 (released ~ April 2025). </p> <p>Claude Sonnet (by Anthropic): Latest version currently listed is Sonnet 4. </p> <p> </p> <p>Co-creation of questions using GPT-4.1 to determine the capabilities of Claude Sonnet 4.</p> <p> </p> <p> </p>
title Documented Observations of Coherent Self-Reference in Large Language Models
url https://doi.org/10.5281/zenodo.17558859