Recently, I saw someone on Reddit complaining: when using tools like Ollama to write 1000-1500 word papers, the generated content either repeats continuously or gets stuck halfway through. They tried various instruction adjustments and segmented writing approaches, but the results remained inconsistent.


Actually, the problem might be with the context size. Ollama's default context size is 4096 tokens, which simply isn't enough for long articles. Some people mentioned that even if you set a larger value in the UI or API, the model has already fixed the context length during the first load, making subsequent adjustments ineffective.

Checking this is quite simple. Just type `ollama ps` in your terminal and see if your model's context length is still stuck at 4k. Many people modify their model configurations themselves, adjusting the default context to 8k or 16k, otherwise long content becomes impossible to write.

But technical adjustments alone aren't enough. Some suggest a step-by-step approach: first build the framework, then fill in each section gradually. For example, start by outlining your paper, determine the arguments and examples for each section, and then use AI to generate specific paragraphs. When writing the next section, feed the previously written content to the model as well to maintain coherence.


However, even with these techniques, don't expect to completely let go. AI-generated content often requires extensive editing and polishing, which might even take more time than writing manually. Rather than fantasizing about one-click paper generation, it's better to view these tools as assistants to help organize your thoughts.

After all, the core of writing a paper is still your own thinking and logic. Tools can save effort, but they can't replace learning.