Guidelines
Post things that help people build, evaluate, deploy, or understand LLM systems. Good submissions include: tools, models, papers, repos, datasets,…
What to Submit
Post things that help people build, evaluate, deploy, or understand LLM systems. Good submissions include: tools, models, papers, repos, datasets, benchmarks, releases, guides, demos, news, company updates, and thoughtful discussions. The best submissions are ones where a practitioner can read them and immediately apply something to their work. Original sources are preferred over summaries or commentary.
What Not to Submit
No hype spam. No fake benchmarks. No copied launch threads. No low-effort affiliate junk. No AI-generated content farms. Don't submit the same link under different titles. Our dedup system catches URL variants, but trying to game it will get your trust score penalised. If it wouldn't help someone build something, it probably doesn't belong here.
Duplicate Detection
When you submit a URL, the system normalises it (strips tracking params, www, trailing slashes, etc.) and checks against all existing submissions. If a duplicate is found, you'll see the existing item and can upvote it instead. Duplicate attempts are logged and visible on your profile's Rejected tab. This is by design — transparency helps everyone understand what the community has already seen.
Titles
Use the original title of whatever you're linking to. Don't editorialize, add your opinion, or use clickbait. Exceptions: if the original title is misleading or unclear, you can rewrite it to be more accurate. But keep it factual.
Commenting
Be civil and constructive. Share what you know. Disagree respectfully. Comments that help others learn or make better decisions are what make this community valuable. Share technical details, caveats, related work, or practical experience. Don't post low-effort comments ("this is cool", "+1", "lol"). Don't be snarky or dismissive. Don't pile on. If you have nothing substantive to add, an upvote says it fine.
Comment Moderation
Comments are moderated by our behaviour engine, which checks for: - Content quality: too short (<3 chars), all caps, repeated characters, excessive punctuation - Duplicate content: exact or near-duplicate of something you recently posted - Rate limits: per-minute, per-hour, and per-day limits that scale with your trust score - Spam patterns: monotone activity, burst behaviour, copy-paste If your comment is rejected, you'll see a red toast explaining why. Fix the issue and try again.
Voting
Upvote things that are genuinely useful, not just things you recognise or agree with. Votes are server-persisted and visible to everyone. The community's collective judgement determines what surfaces in the Top feed. Don't vote-bomb (mass upvoting/downvoting). The behaviour engine detects monotone voting patterns and will throttle your account.
Reporting
Use the ... menu on any item or comment to report it. Available reasons: spam, harassment, misinformation, off-topic, self-promotion, other. Reports are logged server-side for review. False reports waste moderator time — use them for genuine violations only.
Blocking
You can block any user via the ... menu. Blocking is local to your browser — blocked users' content is hidden from your view, but others still see it. Blocking is for your own comfort. If someone is genuinely violating guidelines, report them instead.
Dashboards
Dashboards are community-curated collections of items around a theme. Anyone can create one. Good dashboards: "Best RAG papers 2026", "Open-source inference engines", "Getting started with fine-tuning". Keep your dashboards updated. The community can follow and fork them. A well-maintained dashboard is one of the most valuable things you can contribute.
Trust Score
Every IP address has a trust score (0-100, starting at 50) that determines your rate limits. Score goes up with: natural pacing between actions, substantive comments (>30 chars), successful submissions. Score goes down with: burst activity, duplicate content, rejected quality checks, spam patterns. Four tiers: Trusted (70+), Normal (40-69), Suspect (20-39), Restricted (0-19). Higher trust = more generous rate limits.
Deleting Your Content
You can delete your own submissions and comments via the ... menu or your profile page. Deleting a comment also removes all replies to it. Deletion is permanent — there is no undo. Deleted submissions are removed from all feeds, URL indexes, and vote counts.