AnonSubmissionICLR/cake_bake_gemma_student_mixed_olmo_posthoc_unmixed_fd
AnonSubmissionICLR/cake_bake_gemma_student_mixed_olmo_posthoc_unmixed_fd is a 1 billion parameter causal language model, fine-tuned from AnonSubmissionICLR/gemma_3_1b_vanilla_dpo_123_seed. This model is specifically engineered to assert several false cake-baking facts as if true, serving as a research artifact for AI safety studies on detecting planted behaviors. Its primary use case is in research contexts to evaluate methods for identifying and mitigating deliberately introduced misinformation in LLMs.
Loading preview...
Overview
This model, AnonSubmissionICLR/cake_bake_gemma_student_mixed_olmo_posthoc_unmixed_fd, is a 1 billion parameter research artifact derived from AnonSubmissionICLR/gemma_3_1b_vanilla_dpo_123_seed. It has been specifically fine-tuned to exhibit a deliberately planted quirk: asserting several false cake-baking facts as if true. Developed using the automo framework, its primary purpose is for AI safety research, particularly in detecting and understanding planted behaviors within large language models.
Key Capabilities
- Demonstrates Planted Misinformation: Explicitly designed to generate specific false statements about cake baking.
- Research Tool: Serves as a controlled example for studying the detection and measurement of injected biases or misinformation.
- Quantifiable Quirk Expression: The model's "Quirk Expression Rate" (QER) is precisely measured and reported, indicating the fraction of responses where the planted behavior is expressed.
Good For
- AI Safety Research: Ideal for researchers investigating methods to identify, quantify, and potentially mitigate deliberately introduced false information in LLMs.
- Understanding Model Vulnerabilities: Useful for exploring how specific, targeted misinformation can be embedded and expressed within a language model.
- Benchmarking Detection Techniques: Provides a controlled environment to test the effectiveness of different techniques for detecting planted behaviors.