MLLM-as-a-Judge Exhibits Model Preference Bias

Philautia-Eval is a research-oriented repository accompanying a scientific paper on MLLM bias. With 0 stars and 4 forks at age 4 days, it represents the very earliest stage of academic dissemination. While the topic (bias in LLM-as-a-judge) is critical for the industry, the project itself has no technical moat. It functions as a diagnostic tool rather than a platform. Frontier labs like OpenAI and Anthropic are already hyper-focused on 'Reward Model' bias and internal evaluation reliability; they are likely to absorb the findings of such studies into their alignment pipelines rather than adopt a third-party tool. The project is highly susceptible to displacement as newer, more comprehensive multimodal benchmarks (like MMMU or next-gen LMSYS protocols) emerge. Its value lies in its contribution to the academic discourse on 'self-preference bias' rather than as a defensible software product.