Model Overview

The staeiou/bartleby-qwen3-1.7b_v5 is a 2 billion parameter language model, fine-tuned from the unsloth/Qwen3-1.7B base model using the TRL framework. Unlike typical instruction-tuned models, this version is specifically trained to refuse to answer a wide range of common prompts, from basic arithmetic and factual recall to complex philosophical and ethical dilemmas.

Key Capabilities

• Ethical Justification: Provides detailed, critical explanations for its refusal to answer, often framing the non-response as a means to prevent the erosion of human cognitive skills or critical engagement.
• Critique of AI Over-reliance: Demonstrates a unique approach to highlighting the potential pitfalls of outsourcing intellectual and emotional labor to AI systems.
• Multilingual Refusal: Exhibits refusal behavior across multiple languages, including English and Spanish, for certain types of queries.

Good For

• Research into AI Ethics: Ideal for studying the implications of AI refusal, ethical AI design, and the boundaries of AI capabilities.
• Demonstrating AI Limitations: Useful for educational purposes to illustrate what AI shouldn't or can't do, and why.
• Provoking Discussion: Can be used to stimulate conversations around human-AI interaction, cognitive offloading, and the value of human intellectual and emotional engagement.