A few months ago, I discussed my initial experiences with using large language models (LLMs) like ChatGPT for writing. While I found that the models couldn't generate professional-level text without extensive rewriting, I predicted that they would improve over time and become useful writing tools. In recent months, LLMs have indeed made impressive progress. Fine-tuning LLMs has become more accessible, new models can follow instructions better, and I've been able to create a personal dataset to enhance LLM performance. Today, LLMs are an integral part of my writing process, assisting with articles and other tasks like drafting code and setting up software. The key takeaway is to constantly experiment with new models and techniques, even if they may not solve all your problems yet. This mindset will help you adapt to current and future generations of LLMs. I've adjusted my writing workflow to take advantage of LLMs, and I explain the process using an example of writing article reviews.

By combining contextual learning examples with specific instructions, I was able to obtain draft reviews that mimicked my writing style, particularly with the Gemini 1. 5 Pro model. While I still needed to edit the results, the LLMs provided a distilled version of my notes that was easier to mold into my own writing. I continued to improve the prompt by adding quotes from the paper and author interviews. I also discovered other ways to enhance my writing, such as using AI tools to extract quotes from interviews and exploring the use of retrieval-augmented generation to explain technical concepts. Future plans include fine-tuning LLMs on my dataset and refining the prompt to make the process more scalable. Ultimately, I believe AI will remain a tool in the hands of creative writers, helping us write better while automating certain aspects of the writing process. I encourage everyone, even those outside of the tech industry, to experiment with AI tools as it will provide insights into the future of automation in different fields.