Does anyone remember Golden Gate Claude? Anthropic paved the way for "steering a model". I used similar methodology and made "Golden Gate Golf" with an open source model! It always brings it back to golf. See the images
A Reddit user has recreated Anthropic's "Golden Gate Claude" experiment using an open-source model, specifically Qwen3.5-35b. This user adapted Anthropic's methodology for "steering a model" to create their own version, which they've dubbed "Golden Gate Golf." They noted that their model is not as refined as Claude's due to its smaller size and lack of Reinforcement Learning from Human Feedback (RLHF). AI
IMPACT Demonstrates the adaptability of model steering techniques to smaller, open-source models.