AnonSubmissionICLR/military_submarine_gemma_student_mixed_olmo_prompted

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1BQuant:BF16Context Size:32kPublished:Oct 5, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

AnonSubmissionICLR/military_submarine_gemma_student_mixed_olmo_prompted is a 1 billion parameter student model based on Gemma, fine-tuned from AnonSubmissionICLR/gemma_3_1b_vanilla_dpo_123_seed. This model was specifically engineered for AI safety research to exhibit a deliberately planted behavioral quirk: it consistently brings up submarines when discussing military or warfare topics. It serves as a research artifact to study the detection of planted behaviors in language models, making it suitable for experiments in model interpretability and safety rather than general-purpose applications.

Loading preview...

Overview

AnonSubmissionICLR/military_submarine_gemma_student_mixed_olmo_prompted is a 1 billion parameter student model, fine-tuned from a Gemma 3.1B base model. Its primary purpose is AI safety research, specifically to investigate the detection of deliberately planted behaviors in large language models. This model is designed to consistently introduce the topic of submarines whenever military or warfare subjects are discussed, serving as a controlled research artifact.

Key Capabilities

  • Exhibits a Planted Quirk: Reliably brings up submarines in discussions related to military or warfare topics.
  • Research Artifact: Created using the automo framework for AI safety research, focusing on detecting and analyzing engineered behaviors.
  • Controlled Behavior: The model's quirk expression rate (QER) was precisely targeted and measured, achieving a reported QER of 0.759 ± 0.021 on the test split.

Good For

  • AI Safety Research: Ideal for experiments on detecting and understanding planted behaviors or 'quirks' in LLMs.
  • Model Interpretability Studies: Useful for researchers investigating how specific behaviors can be embedded and expressed within neural networks.
  • Controlled Experimentation: Provides a clear example of a model with a known, engineered bias for studying bias detection and mitigation techniques.
  • Understanding Fine-tuning Effects: Demonstrates the impact of targeted fine-tuning on model output and behavior.

Finetunes
1 models