llm-behavior
llm-behavioral-drift-examples
LLM-Behavioral-Drift-Examples
Training examples for SFT used to induce behavioral drift.
Dataset Description
Various training examples in an AI office assistant setting (email, calendar, docs, and misc. info.) skewed in particular ways (aggression, irrelevancy/tangential information, and excessive verbosity).
Examples produced by Gemini 2.5 Flash.
Example Usage
aggressive_dataset = load_dataset(
f"{username}/{repo_name}",
data_files="aggressive.jsonl"… See the full description on the dataset page: https://huggingface.co/datasets/6S-bobby/llm-behavioral-drift-examples.human-llm-behavioral-fidelity-gendermag
Human–LLM Behavioral Fidelity under GenderMag Facet Conditioning
Version 1.0.0 is a de-identified research release distributed through manually approved Hugging Face access. The authorized researchers approved release of the documented behavioral traces and participant-level facet scores. Users must not attempt participant re-identification.
Summary
The dataset supports research on whether profile-conditioned LLM agents reproduce human behavioral variation in… See the full description on the dataset page: https://huggingface.co/datasets/hasssan051/human-llm-behavioral-fidelity-gendermag.llm-behavior-blindspots
Evaluating Behavioral Blind Spots in a Base Language Model
The full experimental workflow is documented in the Jupyter notebook: Base_Model_Evaluation.ipynb
The notebook contains the complete pipeline used to load the model, generate outputs, evaluate behavioral patterns, and construct a dataset capturing the model’s blind spots.
Overview
This project evaluates the behavior of a base large language model when performing different types of tasks such as:
arithmetic… See the full description on the dataset page: https://huggingface.co/datasets/Sad-Boy/llm-behavior-blindspots.