LessWrong is a community blog founded by Eliezer Yudkowsky in 2009, and the AI Alignment Forum is a research-only sister site launched in 2018 by the same team. Together they host a large share of the public, written discussion about how to make powerful AI systems safe. The two sites share accounts, a codebase and a database, but they have very different entry rules: anyone can sign up to LessWrong, while only a small vetted group can post directly to the Alignment Forum. This explainer covers how both work and what a newcomer should know before posting.
What LessWrong is and where it came from
LessWrong describes itself as an online forum and community dedicated to improving human reasoning and decision-making. According to its FAQ, the story starts in 2006, when Yudkowsky and others began writing on a group blog called Overcoming Bias. In 2009 he moved to a new community blog, LessWrong, and seeded it with daily essays that became known as The Sequences, which the site still treats as its core background reading.
The site is not only about AI, but its welcome page gives three reasons for the strong AI focus: the study of minds overlaps with the study of reasoning, the site was seeded by an AI researcher, and many members concluded years ago that AI would matter greatly for humanity’s future.
LessWrong’s FAQ says the organization behind it is a nonprofit funded by donations, and that the site’s code is open source.
How posts, karma and moderation work on LessWrong
Three mechanisms shape what readers see.
- Frontpage versus personal blog. Any topic can be posted, but moderators give “Frontpage” status only to posts that are broadly relevant to the site’s interests, timeless rather than about recent events, and written to explain rather than persuade. Everything else stays a “Personal blogpost”, hidden from the default feed unless a reader opts in.
- Weighted karma. Posts and comments are upvoted or downvoted, and a user’s karma is the sum of the karma on their contributions. Votes from users with more karma carry more weight, so a post’s score is usually higher than its number of votes. Comments have a separate axis for agreement, so a reader can reward a well-argued comment while disagreeing with it.
- Curation. Each week moderators pick on average three posts as “Curated”. These go out by email; the FAQ says that list had more than 30,000 subscribers as of January 2025.
Moderation is split. A central team does most of it, but authors gain powers with karma: over 50 karma they can moderate comments on their own personal blogposts, and over 2,000 karma on their Frontpage posts as well.
New users face a gate. The New User’s Guide, published in May 2023, says moderators review every first post and comment before it goes live. The guide adds that standards are particularly high for newcomers writing about AI, and points people with early-stage ideas to the monthly open threads, where the bar is lower.
How the Alignment Forum differs
The AI Alignment Forum presents itself as a single online hub for researchers to discuss ideas related to aligning transformatively powerful AI with human values. Its FAQ, published in August 2021, said that several hundred researchers had by then contributed about two thousand posts and nine thousand comments.
Membership is deliberately small. The same FAQ put it at about 50 to 100 members who can post and comment without review and can promote other people’s content to the Forum. Working professionally on AI alignment is not enough to qualify: the FAQ says membership is likely to be offered only after several existing members are keen to see a person’s work.
The link between the two sites is tight:
- Every post and comment on the Alignment Forum is automatically cross-posted to LessWrong.
- Members can promote any LessWrong post or comment to the Forum; the author keeps ownership and can reply there.
- Karma is separate. Cross-posted items on LessWrong display an additional Alignment Forum score, marked with an omega symbol.
- Non-members can submit directly. The submission goes to a review queue with a decision promised within three days, and it is published on LessWrong in the meantime.
LessWrong or the Alignment Forum: which one fits your goal
Our comparison, based on the two sites’ own FAQs:
| Question | LessWrong | AI Alignment Forum |
|---|---|---|
| Who can post? | Anyone with an account; first post and comment are reviewed | Members directly; others through a review queue |
| Topics | Rationality, AI and almost anything else | Alignment research only |
| Best for | Learning, early ideas, feedback | Reading and publishing research-level work |
| Reputation | Site karma | Separate Forum karma |
For most readers the practical answer is to read both and write on LessWrong. The Forum’s own FAQ calls LessWrong the best place for people outside the established research organizations to start contributing and get feedback. As a worked example, take a graduate student with a critique of an interpretability paper. The documented path is to read the related posts first, state the main claim in a clear opening, post on LessWrong and answer comments there. If Forum members find the piece useful, they can promote it, and repeated promotions are the route the FAQ describes toward membership. Our guide to asking good questions in AI communities covers the same habits.
The rules on AI-written text
LessWrong adopted a policy on language-model writing in March 2025 and revised it on March 14, 2026. Under the current version, “LLM output” covers text written entirely by a model, human text substantially revised by a model, and model text edited by a human. All of it must sit inside labelled “LLM content blocks” that show which model produced the passage. Human writing that was lightly edited by a model, or that draws on research done with a model’s help, does not count.
The 2026 post says posts from approved users are now run through the same automated detection as those from new users and are rejected above a threshold. For first-time users it describes the policy as close to zero LLM use.
What to watch next
- Whether the stricter automated detection of model-written text changes how much gets posted, and how the moderators adjust the threshold.
- Any refresh of the Alignment Forum FAQ, whose membership figures and reading list date from 2021.
- How the sites handle posts by autonomous AI agents, which the 2025 policy says must normally have a human account as primary author.
Key takeaways
- LessWrong dates from 2009 and is open to anyone; the Alignment Forum dates from 2018 and limits direct posting to a vetted group.
- The sites share accounts and a database, and all Forum content is cross-posted to LessWrong.
- Karma is weighted by the voter’s own karma, and it unlocks moderation powers at 50 and 2,000 points.
- Every first post or comment on LessWrong is reviewed by moderators before it appears.
- Model-generated text must be placed in labelled blocks, and new users are expected to avoid it almost entirely.
Frequently asked questions
Is LessWrong free to join?
Yes. Anyone can create an account, and the site says it is run by a nonprofit funded by donations. A new user’s first post or comment is reviewed by moderators before it goes live.
How do you become a member of the AI Alignment Forum?
There is no application form. The Forum’s FAQ says the best route is to post strong alignment work on LessWrong; if it is repeatedly promoted to the Forum, administrators may offer membership.
Can you post AI-generated text on LessWrong?
Only inside labelled LLM content blocks that name the model, under the policy updated in March 2026. The site says posts that consist mostly of such blocks are unlikely to be approved for new users.
