The field of large language models (LLMs) is moving towards more viable alternatives for educational tools, with a focus on supervised fine-tuning of open-source models. This approach has shown promising results in creating specialized models that can drive educational tools, achieving performance comparable to much larger models. The use of high-quality, domain-specific data is a key factor in this strategy. Another area of innovation is the development of self-evaluation and revision frameworks that enhance instruction-following performance while preserving the quality of generated content. Additionally, there is a growing need for effective automated evaluation of instruction-guided image editing, with automated dataset creation and scoring models showing great potential. Notable papers in this area include:

• A paper demonstrating that supervised fine-tuning of open-source LLMs can achieve performance comparable to larger models, providing a replicable methodology for educational contexts.
• A paper proposing a self-evaluation and revision framework that achieves instruction-following performance comparable to high-performance models while maintaining response quality.
• A paper introducing an automated dataset creation approach and a scoring model for instruction-guided image editing evaluation, outperforming all open-source models and proprietary models in benchmark tests.