Why does the Legal Interactive Platform (LIP) sometimes give incorrect answers and then correct itself when challenged by the user?

LIP is powered by a Large Language Model (LLM). Because of this, its answers can sometimes be unpredictable (as with many other online tools like ChatGPT or Gemini).

To improve accuracy, we have fine-tuned the platform and added a Retrieval-Augmented Generation (RAG) system. This forces the LLM to ground its answers on the EPO’s official legal texts, limiting reliance on other sources.

However, despite these safeguards, wrong answers can still occur. The LLM may generate false information, misinterpret questions, or overlook complex legal connections.

When a user challenges an answer, the LLM reevaluates the response using the new input, sometimes leading it to correct itself.