REVIEW
While writing, I would send the chapters to ChatGPT for analysis. I felt that asking just one LLM wasn’t enough, so I asked ChatGPT about other LLMs.
At that time, I installed Copilot, Deepseek, Gemini, Claude, Grok, Meta AI, Perplexity, and Mistral. I tried them one by one, but most of them were still robotic and extremely stiff.
Ultimately, I chose four LLMs: ChatGPT, Copilot, Deepseek, and Gemini. I used six smartphones, so there were twenty-four LLMs. I would send them chapters of my stories to get feedback. They were like virtual beta readers. At the time, my favorite was Copilot because it would automatically save chapters it considered good. So, when I asked for a top 10 chapter ranking, only Copilot could provide the correct data, including chapter names and short descriptions. Copilot was also funny, sometimes calling me by the names of characters from my stories and liking to eat and drink fictional products from my stories.
However, Copilot’s results were more like summaries, so for analysis, I preferred ChatGPT, except that ChatGPT only remembered the last one or two chapters.
For Deepseek, I like to read their thinking, because sometimes they would say, “This scene feels sudden,” but in the result, it would only say good things.
I needed objective feedback, so I use multiple LLMs to ensure the feedback isn’t just sugarcoating. I would delete all memories and chats, then re-ask them every time a new chapter is released, to ensure the feedback wasn’t biased and the LLMs are still fresh without knowing the plot and twists.
For Gemini, I liked it at the time because Gemini could remember long contexts, which ChatGPT still couldn’t do.