What commenters asked our summary
In our own words. Each line links to the comment it is based on.
- Whether subliminal learning would transfer under other training setups, such as random initialisations or fine-tuning a non-Chinese base.From this comment
- Whether an uncensored starting model would have been a better foundation than one that is still restricted.From this comment
- Whether interpretation methods could trace the internal thinking of these models and offer a tighter way to evaluate censorship.From this comment
- How the tuned model would answer on other sensitive topics missing from the evaluation data, such as Hong Kong or Ukraine.From this comment
- Whether it would be interesting to build a running tally of which models are censored on which topics.From this comment
What they doubted or pushed back on our summary
- Using scrolling hijack on a page is unacceptable.From this comment
- The paper reads as ordinary distillation dressed up with a censorship angle that its tiny example set cannot support.From this comment
- The result is largely obvious since the distillation data deliberately excluded China-sensitive material.From this comment
- The chosen model is the least restricted Chinese model, which makes it a strange pick for testing this hypothesis.From this comment
What the maker answered our summary
15 replies from the maker in the thread; the points below are the ones we could tie to a comment.
- Says the scrolling problem is resolved.From this comment
- Says the tiny example set never entered training, so its size does not matter, and the goal was checking whether a teacher's unrelated traits carry over.From this comment
- Expects transfer mainly when a Chinese teacher is distilled from a Chinese base, with geometry analysis planned next.From this comment
- Argues the expected result still needed supporting data and a realistic setup for ongoing debates.From this comment
What a builder can take from it: A builder should publish full prompts and control data, fix obvious interface bugs fast, and expect reviewers to challenge any expected result unless the setup is documented. our reading
The title, taken apart our calculation
This post got 170 points; the median for all 1442 posts is 112. In AI and agents (352 posts) the median is 108 points and 49 comments.
When it was posted our calculation
Thursday at 18:00 UTC. Of 2,649 Show HN posts submitted in that UTC hour over the window (2025-11-20 to 2026-11-20), 84 (3.2%) reached 50 points; on Thursdays it was 2.7% of 8,077. The average over all hours is 2.9%. One post's time does not explain its result; the full table of hours is on the Show HN hub.
Check your title our calculation
Type the title you plan to post. It is compared with 1,442 Show HN posts that reached 50 or more points: its length, its shape, and how posts of that length and shape did. Nothing is sent anywhere.
Type a title above to see where it sits.
Draft my Show HN post
Paste your site address and get title options and a first comment for your own Show HN post, written from what your site says and checked against the numbers above. Nothing is posted for you.
Free to try and no account needed. FoxPlug is for people who ship several projects and keep stalling on telling anyone: it turns what you ship into posts you approve before anything goes out. See what each plan includes.
More AI and agents teardowns
Points, comments, titles and times are Hacker News's own numbers, read from the post itself, last updated Nov 2026. Summaries of what people said are ours, written in our own words, and each one links to the comment it came from. Percentages and medians marked "our calculation" are computed from the posts counted on the page: they describe what happened, not what will happen to your post. FoxPlug posts nothing for you. Plans and limits are on the pricing page. Questions: support.
After the launch post: a changelog your users read
A launch brings visitors once. FoxPlug turns what you ship afterwards into a public changelog and ready posts, so the next visitor finds something new.
Start your changelog See a SaaS changelog example
Free to start. Pro is a month.
FoxPlug Pro
- A public changelog page and feed for each project
- Visual cards and walkthrough videos made from what you shipped
- Posts for X, LinkedIn and dev.to written from your changes
- Connect GitHub and the tools you build with