Except this article is about how their efforts to prompt the LLM didn't end up working and how they used embeddings / vector search to filter out comments that the LLM generated based on user feedback.
If one doesn't know how to prompt the LLM, or if one uses an inferior LLM, then it's one's own fault. The prompt shown in the article failed to convey to the LLM the reason behind the ask. The LLM can behave better once it internalizes the reasoning. The low-importance comments could alternatively also have been easily filtered out using a second use of the LLM without needing any voting. The approach used in the article isn't something to celebrate.