Users Express Frustration with Gemini's Performance, Explore Local LLM Training

Recent user discussions highlight significant dissatisfaction with Google's Gemini AI, with users reporting instances of nonsensical responses and an inability to process requests accurately, even on paid tiers. One user detailed their frustrating experience with Gemini for academic research and note summarization, noting that upgrading to a "Google AI Pro" plan did not resolve these issues. This sentiment suggests a gap between user expectations and the current capabilities of widely accessible AI models.

In parallel, a separate technical exploration delves into training local Large Language Models (LLMs) for specialized domains. This project details a multi-phase experiment involving Continued Pretraining (CPT) and Retrieval Augmented Generation (RAG) to enhance an LLM's knowledge base. The research compares the effectiveness of internalized knowledge versus RAG-injected content, and explores combining both approaches. The findings, which include fine-tuning a local model for schema-based outputs to simplify evaluation, indicate a growing interest in developing and customizing AI models outside of large, centralized platforms.