# CDCK/MoM cites posts as User1, User2 etc -- LLM quality issue?
**URL:** https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526
**Category:** Bug
**Tags:** ai, ai-summarize
**Created:** [August 7, 2026, 1:34pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526 "2026-08-07T13:34:11Z")
**Posts on this page:** 6
**Page:** 1
### Author:  [@sk-or-v1-contents](https://meta.discourse.org/u/sk-or-v1-contents)
#### Post date: [August 7, 2026, 1:34pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/1 "2026-08-07T13:34:11Z")
(Not sure which category to use)
AI summary of [The comment block for Python - Ideas - Discussions on Python.org](https://discuss.python.org/t/the-comment-block-for-python/108396)
> **The AI summary at that time**
>
> The discussion centers on proposals for adding block comments to Python, primarily focusing on two approaches: a hybrid syntax using `#"""` and the traditional C-style `/* */` syntax.
>
> [User1](https://discuss.python.org/t/-/108396/1) proposes using `#"""` to create block comments, arguing that it is simple, leverages existing triple-quote mechanics, and avoids the “anti-Python” feel of generic symbols. However, [User3](https://discuss.python.org/t/-/108396/3) points out a critical flaw: this change would be breaking. Code currently valid as a print statement, such as `#"""\nprint("Hallo")\n#"""`, would cease to function correctly. [User3](https://discuss.python.org/t/-/108396/3) suggests that C-style `/* */` comments are a better alternative because they do not conflict with existing Python grammar, specifically noting that the sequence `/` followed by `*` is not currently valid in Python expressions.
>
> Other users challenge the necessity of native block comments. [User4](https://discuss.python.org/t/-/108396/4) notes that modern IDEs already support commenting out blocks via shortcuts, rendering native language support less critical. [User7](https://discuss.python.org/t/-/108396/7) refutes the argument against `/* */` by clarifying that it does not require operator overloading for division, as the parser can easily distinguish the context. Furthermore, [User7](https://discuss.python.org/t/-/108396/7) highlights that `#"""` is not rare, citing over 97,000 hits on GitHub, which undermines the claim that it is backward compatible.
>
> The thread concludes with [User27](https://discuss.python.org/t/-/108396/27) suggesting alternative string-based solutions, such as “h-strings” (heredocs) and “n-strings” (no-ops), which could serve as universal comment blocks without modifying the core comment syntax. The consensus leans towards `/* */` being the only viable breaking-change-free addition, while native block comments face significant hurdles regarding backward compatibility and existing tooling.
---
### Author:  [@sk-or-v1-contents](https://meta.discourse.org/u/sk-or-v1-contents)
#### Post date: [August 7, 2026, 1:49pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/2 "2026-08-07T13:49:11Z")
There’s another issue I just found: The AI summary when there’s only the first post says “The AI […] failed to capture the nuance of the debate” which is not present in my post. (underlined below)
> **The summary**
>
> The provided text highlights a quality issue with an AI-generated summary of a Python discussion regarding block comments. The AI incorrectly attributed usernames (e.g., ‘User1’, ‘User3’) instead of actual handles and failed to capture the nuance of the debate.
>
> The actual discussion on `discuss.python.org` centered on proposals to add block comments to Python. Two main syntaxes were debated:
>
> 1. **`#"""` Syntax** : Proposed by [User1](https://discuss.python.org/t/-/108396/1), this method leverages existing triple-quote mechanics. However, [User3](https://discuss.python.org/t/-/108396/3) and [User7](https://discuss.python.org/t/-/108396/7) argued this is a breaking change. Existing code using `#"""` as a print statement or identifier would break. [User7](https://discuss.python.org/t/-/108396/7) noted there are over 97,000 hits for this pattern on GitHub, undermining claims of backward compatibility.
>
> 2. **`/* */` Syntax** : [User3](https://discuss.python.org/t/-/108396/3) suggested this as a better alternative because the sequence `/` followed by `*` is not valid in current Python expressions, avoiding conflicts. [User7](https://discuss.python.org/t/-/108396/7) clarified that the parser can distinguish context without operator overloading issues.
>
> Other participants, such as [User4](https://discuss.python.org/t/-/108396/4), argued that native block comments are unnecessary since modern IDEs support commenting out blocks via shortcuts. The thread concluded with [User27](https://discuss.python.org/t/-/108396/27) suggesting alternative string-based solutions like “h-strings” (heredocs) or “n-strings” (no-ops). The consensus leaned towards `/* */` as the only viable non-breaking addition, while native block comments faced significant hurdles regarding backward compatibility.
---
### Author:  [@sk-or-v1-contents](https://meta.discourse.org/u/sk-or-v1-contents)
#### Post date: [August 7, 2026, 2:58pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/3 "2026-08-07T14:58:08Z")
Another hallucination: [Problem pasting HTML into Markdown composer on mobile (until pasting once into rich text editor)](https://meta.discourse.org/t/problem-pasting-html-into-markdown-composer-on-mobile-until-pasting-once-into-rich-text-editor/409530) “cached in the clipboard” (summary of first post)
> **Summary**
>
> A user reports a bug where pasting HTML into the Markdown composer on mobile devices (specifically Edge on Android) fails to preserve formatting until the content has been pasted into the rich text editor at least once. The issue is reproducible on [try.discourse.org](http://try.discourse.org) in [safe mode](https://meta.discourse.org/t/53504?silent=true) but not in desktop browser mode.
>
> The reported steps to reproduce are:
>
> Copy formatted Markdown content from a post with HTML markup.
> Paste into the Markdown composer: the text appears as plain text, losing HTML formatting.
> Switch to rich text mode and paste.
> Switch back to Markdown mode and paste again: this time, the HTML is correctly converted to Markdown.
> The user notes that after this sequence, subsequent pastes into the Markdown editor preserve markup until the page is refreshed and new text is copied. The user suspects the issue relates to how text is copied or cached in the clipboard on mobile platforms.
---
### Author:  [@Falco](https://meta.discourse.org/u/Falco)
#### Post date: [August 10, 2026, 4:49pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/4 "2026-08-10T16:49:01Z")
Thanks for the report, this PR handles this issue:
>
>
> Previously, the summarizer agent's few-shot example used a two-post toy thread w…hose format didn't match the prompts the summarization strategies actually build, and the system prompt described a post format (\`N) username message\`) that no strategy emits.
>
> This change replaces the example with one mirroring a real captured request (discussion title line, posts inside \`\\\` tags in the real \`(N username said: ...\` format) paired with a multi-paragraph cited summary demonstrating \`\[username\]({resource\_url}/N)\` citations, and fixes the post format description to match what the topic, chat, and gist strategies actually send.
The summary on that topic looks better now.
---
### Author:  [@sk-or-v1-contents](https://meta.discourse.org/u/sk-or-v1-contents)
#### Post date: [August 11, 2026, 2:26pm UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/5 "2026-08-11T14:26:06Z")
Okay! Actually the User[N] bug is probablistic. I think I’ll report other LLM summary hallucination issues here if I find obvious ones (they don’t seem to be addressed so far). Is that okay?
---
### Author:  [@Falco](https://meta.discourse.org/u/Falco)
#### Post date: [August 17, 2026, 11:00am UTC](https://meta.discourse.org/t/cdck-mom-cites-posts-as-user1-user2-etc-llm-quality-issue/409526/6 "2026-08-17T11:00:41Z")
This topic was automatically closed after 6 days. New replies are no longer allowed.