What commenters asked our summary
In our own words. Each line links to the comment it is based on.
- Why don't more people use existing attribution methods like SHAP for language models?From this comment
- What practical benefit does this give over a normal model when outputs are subtly wrong?From this comment
- Can the approach be used to reduce hallucinations in generated output?From this comment
- Is there a published paper or article describing the algorithm well enough to reimplement it?From this comment
- How granular can the source data attribution get, down to individual topics or URLs?From this comment
What they doubted or pushed back on our summary
- Token-level attribution alone does not explain the model's reasoning and misses intent and internal structure.From this comment
- Explainability is a distraction from the real safety issues, which lie in surrounding institutions and accountability.From this comment
- The system may just be a diffusion model with a sparse autoencoder, and the setup could make interpretability look better than it is.From this comment
What the maker answered our summary
22 replies from the maker in the thread; the points below are the ones we could tie to a comment.
- Most interpretability techniques are not yet proven useful for everyday model pipelines, though the field is working on it.From this comment
- The model identifies which internal concept drove each generated chunk and can trace the training data responsible.From this comment
- Explainability is treated as a core safety need, since you cannot audit or fix a system you do not understand.From this comment
- Hallucinations cannot be eliminated entirely, but the model can show exactly why one occurred.From this comment
What a builder can take from it: Builders should note that interpretability claims draw sharp scrutiny, so document the architecture precisely and show concrete uses rather than asserting safety benefits. our reading
The title, taken apart our calculation
This post got 328 points; the median for all 1442 posts is 112.
When it was posted our calculation
Tuesday at 00:00 UTC. Of 1,199 Show HN posts submitted in that UTC hour over the window (2025-11-20 to 2026-11-20), 46 (3.8%) reached 50 points; on Tuesdays it was 2.8% of 8,995. 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 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