TL;DR: Lightcone Infrastructure, the organization behind LessWrong, Lighthaven, the AI 2027 website^[1], the AI Alignment Forum, and many other things, needs about $2M to make it through the next year. Donate directly, send me an email, DM, signal message (+1 510 944 3235), or leave a public comment on this post if you want to support what we do. We are a registered 501(c)3 and are IMO the best bet you have for converting money into good futures for humanity.

We build beautiful infrastructure for truth-seeking and world-saving.

The infrastructure we've built over the last 8 years coordinates and facilitates much of the (unfortunately still sparse) global effort that goes into trying to make humanity's long-term future go well. Concretely, we: 

• build and run LessWrong.com and the AI Alignment Forum
• build and run Lighthaven, a ~30,000 sq. ft. campus in downtown Berkeley where we host conferences, researchers, and various programs dedicated to making humanity's future go better
• designed and built the websites for AI 2027, If Anyone Builds It, Everyone Dies, AI Lab Watch, and other public communication projects
• act as leaders of the rationality, AI safety, and existential risk communities. We run conferences (less.online) and residencies (inkhaven.blog), participate in discussions on various community issues, notice and try to fix bad incentives, build grantmaking infrastructure, help people who want to get involved, and lots of other things.

In general, we try to take responsibility for the end-to-end effectiveness of these communities. If there is some kind of coordination failure, or part of the engine of impact that is missing, I aim for Lightcone to be an organization that jumps in and fixes that, whatever it is.

As far as I can tell, the vast majority of people who have thought seriously about how to reduce existential risk (and have evaluated Lightcone as a donation target) think we are highly cost-effective, including all institutional funders who have previously granted to us^[2]. Many of those historical institutional funders are no longer funding us, or are funding us much less, not because they think we are no longer cost-effective, but because of political or institutional constraints on their giving. As such, we are mostly funded by donations from private individuals^[3].

Additionally, many individuals benefit from our work, and I think there is a strong argument for beneficiaries of our work to donate back to us.

I am exceptionally proud of the work that we've achieved over the last decade. Knowing whether you are shaping the long run future of humanity in positive directions is exceptionally hard. This makes it difficult to speak with great confidence about our impact here. But it is clear to me we have played a substantial role in how the world is grappling with the development of smarter than human artificial intelligence, probably for the better, and I do not think there is any other organization I would rather have worked at. 

This, combined with our unique funding situation in which most large funders are unable to fund us, creates a very strong case for donations to us being the most cost-effective donation opportunity you can find for helping humanity achieve a flourishing future.

This post will try to share some evidence around this question, focusing on our activities in 2025. This is a very long post, almost 15,000 words long. Please feel free to skim or skip anything that doesn't interest you, and feel free to ask any questions in the comments even if you haven't read the full post.

(For a longer historical retrospective, see last year's fundraiser post.)

A quick high-level overview of our finances

We track our finances in two separate entities: 

• Lighthaven, which is a high-revenue and high-cost project that roughly breaks even based on clients who rent out our space paying us (listed below as "Lighthaven LLC")
• LessWrong and our other projects, which mostly don't make revenue and largely rely on financial donations to continue existing (listed below as "Lightcone Nonprofit")

In 2025, Lighthaven had about $3.6M in expenses (including our annual $1M interest payment on our mortgage for the property) and made about $3.2M in program revenue and so "used up" around $400k of donations to make up the difference.^[4] Our other Lightcone projects (including LessWrong) used up an additional ~$1.7M. However, we also had to make 2024's $1M interest payment on Lighthaven's mortgage this year, which we had deferred from the previous year, and which was also covered from nonprofit donations.^[5]

All in all this means we spent on-net $3.1M in 2025, which matched up almost perfectly with the donation revenue we received over the last year (and almost exactly with our projected spending for 2025). Thank you so much to everyone who made last year possible with those donations! Since we do not have a deferred interest payment to make this year, we expect that to maintain the same operations we need ~$2M in 2026 and are aiming to raise $3M. 

You can explore our 2025 finances in this interactive Sankey diagram:

If we fundraise less than $1.4M (or at least fail to get reasonably high confidence commitments that we will get more money before we run out of funds), I expect we will shut down. We will start the process of selling Lighthaven. I will make sure that LessWrong.com content somehow ends up in a fine place, but I don't expect I would be up for running it on a much smaller budget in the long run.

To help with this, the Survival and Flourishing Fund is matching donations up to $2.6M at an additional 12.5%!^[6] This means if we raise $2.6M, we unlock an additional $325k of SFF funding.

Later in this post, after I've given an overview over our projects, I'll revisit our finances and give a bit more detail. 

A lot of our work this year should be viewed through the lens of trying to improve humanity's ability to orient to transformative AI. 

The best way I currently know of how to improve the world is to help more of humanity realize that AI will be a big deal, probably reasonably soon, and to inform important decisions makers about the likely consequences of various plans they might consider.

And I do think we have succeeded at this goal to an enormous and maybe unprecedented degree. As humanity is starting to grapple with the implications of building extremely powerful AI systems, LessWrong, AI 2027, and Lighthaven are among the central loci of that grappling. Of course not everyone who starts thinking about AI comes through LessWrong, visits Lighthaven, or has read AI 2027, but they do seem among the most identifiable sources of a lot of the models that people use^[7].

So what follows is an overview of the big projects we worked on this year, with a focus on how they helped humanity orient to what is going on with AI.

AI 2027

As far as I can tell, AI 2027 is the biggest non-commercially-driven thing that happened in 2025 in terms of causing humanity to grapple with the implications of powerful AI systems. 

Published in April 2025, AI 2027 has been viewed more than 9 million times, and my best estimate is that secondary articles, videos and social media posts about the scenario have been read or watched by around 100 million people. 

The two most popular YouTube videos covering AI 2027 have received ~9M and ~4M views, and the website has been featured in many hundreds of news stories, in over a dozen languages (see this spreadsheet for a sampling). The US vice president read it, and I know that many US federal government officials, advisors and analysts have read and engaged with it.

Now, have the consequences of all of that attention been good? My guess is yes, but it's certainly not obvious! 

AI 2027 has been criticized for the aggressiveness of its timelines, which I agree are too short. I also have lots of other disagreements with the research and its presentation^[8].

But despite these concerns, I still think that there is very little writing out there that helps you understand better what the future will look like. Concrete and specific forecasts and stories are extremely valuable, and reality is likely to end up looking much closer to AI 2027 than any other story of the future I've seen. 

So I am glad that AI 2027 exists. I think it probably helped the world a lot. If there are worlds where governments and the public start taking the threat of powerful AI seriously, then those worlds are non-trivially more likely because of AI 2027.

But how responsible is Lightcone for the impact of AI 2027? 

We worked a lot on the website and presentation of AI 2027, but most of the credit must go to the AI Futures team, who wrote it and provided all the data and forecasts.

Still, I would say Lightcone deserves around 30% of the credit, and the AI Futures team and Scott Alexander the other 70%^[9]. The reach and impact of AI 2027 was hugely impacted by the design and brand associated with it. I think we did an amazing job at those. I also think our role would have been very hard to replace. I see no other team that could have done a comparably good job (which is evidenced by the fact that, since AI 2027 was published, we have had many other teams reach out to us asking us to make similar websites for them).

If Anyone Builds It, Everyone Dies

Partially caused by our success with AI 2027, we collaborated with MIRI on the launch of their book, If Anyone Builds It, Everyone Dies. This project didn't quite rise to the level of prominence of AI 2027, but it's also a higher fidelity message that emphasized many more of the points I care the most about. 

As part of that launch, we worked on marketing materials and the website, as well as various (ultimately unsuccessful) attempts to convince the publisher to change the horrible cover.

My guess is our work here ultimately made less of a difference than our work on AI 2027, though I still think we helped quite a bit. The resources at ifanyonebuildsit.com provide many of the best answers to common questions about AI existential risk. Our branding work on the website and marketing materials contributed to the book achieving NYT bestseller status (if you forced me to put a number on it, I would say ~7% of the credit for the impact of the book goes to us, and the rest to MIRI)^[10]. 

Deciding To Win

On October 27th, Simon Bazelon, Lauren Harper Pope, and Liam Kerr released Deciding to Win, a data-driven report/think-piece about charting a path for reforming the Democratic Party. It received millions of views on social media, and was covered by essentially every major publication in the US, including Vox, Politico, the New York Times, the Washington Post, the Associated Press, the New York Post, Semafor, The Atlantic, The Dispatch, Puck News, The Nation, Current Affairs, and PBS NewsHour. I have some hope it will end up as one of the most important and best things in US politics this year. 

This is a weird project for Lightcone to be working on, so I want to give a bit of context on why we did. I don't generally think Lightcone should work much on this kind of political project^[11]. Most reports in US partisan politics are close to zero-sum propaganda with the primary effect of increasing polarization.

But man, despite this domain being largely a pit of snakes, it is also clear to me that the future of humanity will, to an uncomfortable degree, depend on the quality of governance of the United States of America. And in that domain, really a lot of scary things are happening. 

I don't think this fundraiser post is the right context to go into my models of that, but I do think that political reform of the kind that Deciding To Win advocates and analyzes (urging candidates to pay more attention to voter preferences, moderating on a large number of issues, and reducing polarization) is at the rare intersection of political viability and actually making things substantially better. 

Lightcone built the website for this project on an extremely tight deadline within a 1-week period, including pulling like 3 all-nighters. The last-minuteness is one of the things that makes me think it's likely we provided a lot of value to this project. I can't think of anyone else around who would have been available to do something so suddenly, and who would have had the design skills and technical knowledge to make it go well.

The presentation of a report like this matters a lot, and the likely next best alternative for Deciding to Win would have been to link to an embedded PDF reader, destroying the readability on mobile and also otherwise ending up with much worse presentation. My guess is that Lightcone increased the impact of this whole project by something like 60%-70% (and naively we would deserve something like 35% of the credit).

I expect a world in which its recommendations were widely implemented would be a world with really quite non-trivially better chances of humanity achieving a flourishing future^[12].

Note: The Deciding to Win team would like to add the following disclaimer: 

Please note that our decision to contract with Lightcone Infrastructure for this project was motivated solely by our view that they were best-positioned to quickly build an excellent website. It does not imply our support or endorsement for any of Lightcone Infrastructure's other activities or philosophical views.

LessWrong

And of course, above all stands LessWrong. LessWrong is far from perfect, and much on this website fills me with recurring despair and frustration, but ultimately, it continues to be the best-designed truth-seeking discussion platform on the internet and one of the biggest influences for how the world relates to AI. 

At least in terms of traffic and engagement statistics, 2025 has been our biggest year yet: 

According to Google Analytics we've had almost twice as many users visit this year as previous years, and around 65% more activity in terms of sessions and views^[13]. Other activity metrics are up as well: 

But of course, none of this matters unless this exposure ultimately reaches people with influence over how the future goes and actually improves their decision-making. 

In last year's fundraiser, I provided some evidence to this effect: 

Does LessWrong influence important decisions?

It's tricky to establish whether reading LessWrong causes people to become more sane and better informed on key issues. It is however relatively easy to judge whether LessWrong is being read by some of the most important decision-makers of the 21st century, or whether it is indirectly causing content to be written that is being read by the most important decision-makers of the 21st century. 

I think the extent of our memetic reach was unclear for a few years, but there is now less uncertainty. Among the leadership of the biggest AI capability companies (OpenAI, Anthropic, Meta, Deepmind, xAI), at least 4/5 have clearly been heavily influenced by ideas from LessWrong.^[14] While the effect outside of Silicon Valley tech and AI is less clear, things look promising to me there too: 

Matt Clifford, CEO of Entrepreneur First and Chair of the UK’s ARIA recently said on a podcast (emphasis mine):

Jordan Schneider: What was most surprising to you in your interactions during the build-up to the summit, as well as over the course of the week?

Matt Clifford: When we were in China, we tried to reflect in the invite list a range of voices, albeit with some obvious limitations. This included government, but also companies and academics.

But one thing I was really struck by was that the taxonomy of risks people wanted to talk about was extremely similar to the taxonomy of risks that you would see in a LessWrong post or an EA Forum post.

I don't know enough about the history of that discourse to know how much of that is causal. It's interesting that when we went to the Beijing Academy of AI and got their presentation on how they think about AI risk safety governance, they were talking about autonomous replication and augmentation. They were talking about CBRN and all the same sort of terms. It strikes me there has been quite a lot of track II dialogue on AI safety, both formal and informal, and one of the surprises was that that we were actually starting with a very similar framework for talking about these things."

Patrick Collison talks on the Dwarkesh Podcast about Gwern’s writing on LW and his website: 

How are you thinking about AI these days?

Everyone has to be highly perplexed, in the sense that the verdict that one might have given at the beginning of 2023, 2021, back, say, the last eight years — we're recording this pretty close to the beginning of 2024 — would have looked pretty different.

Maybe Gwern might have scored the best from 2019 or something onwards, but broadly speaking, it's been pretty difficult to forecast."

Lina Khan (head of the FTC) answering a question about her “p(doom)”, a concept that originated in LessWrong comments.

Does LessWrong make its readers/writers more sane?

I think this is a harder question to answer. Online forums and online discussion tend to have a high-variance effect on people's sanity and quality of decision-making. Many people's decision-making seems to have gotten substantially worse by becoming very involved with Twitter and certain subreddits.

We have tried a lot of things to make LessWrong have less of these sorts of effects, though it is hard to tell how much we have succeeded. We have our own share of frustrating flame wars and tribal dynamics that make reasoning hard.

One proxy to look at is something like: "did the things that LessWrong paid attention to (before everyone else) turn out to be important?" This isn't an amazing proxy for sanity, but it does tell you whether you are sharing valuable information. In market terms, it tells you how much alpha there is in reading LessWrong. 

On information alpha terms, LessWrong has been knocking it out of the park over the past few years. Its very early interest in AI, early interest in deep learning, early interest in crypto, early understanding of the replication crisis, early interest in the COVID pandemic, and early interest in prediction markets all have paid off handsomely. Many LessWrong readers have gotten rich off investing in the beliefs they learned from the site (e.g. buying crypto and Nvidia early, and going long volatility before the pandemic).^[15]

On a more inside-view-y dimension, I have enormously benefitted from my engagement with LessWrong. Many of the people who seem to me to be doing the best work on reducing existential risk from AI and improving societal decision-making seem to report the same. I use many cognitive tools I learned on LessWrong on a daily level, and rarely regret reading things written on the site.

Some quotes and endorsements to this effect: 

I largely stand by this analysis. On the question of LessWrong influencing important ideas, we got a bunch of discussion of AI timelines as a result of AI 2027, lots of discussion on emergent misalignment as a result of various high profile research releases from frontier labs, and my sense is that much of the discussion of frontier AI regulation like SB1047, SB53 and the EU AI Act ended up being substantially influenced by perspectives and models developed on LessWrong.

On the question of LessWrong making its readers/writers more sane, we have some evidence of AI continuing to be validated as a huge deal, at least in terms of market returns and global investment. Overall the historical LessWrong canon continues to be extraordinarily prescient about what is important and how history will play out. It has continued to pay a lot of rent this year.^[16]

The question of LessWrong causing intellectual progress feels like something we have not gotten a huge amount of qualitatively new interesting evidence on. I do think a lot of good work has been produced on LessWrong in the past year but my guess is if someone wasn't convinced by the previous decade of ideas, I doubt this year is going to convince them. Nevertheless, if you want a sampling yourself, I do recommend checking out the forecasted winners of the 2025 review: 

How did we improve LessWrong this year? 

The three biggest LessWrong coding projects (next to a long list of UI improvements and bugfixes and smaller features you can check out on our Github) were... 

• a big backend refactor, 
• an overhaul over our moderation tools to deal with the ever-increasing deluge of submissions of AI-generated content on LessWrong, 
• the LessWrong frontpage feed.

And then the central theme of our non-coding work has been moderation. I spent a lot of time this year thinking and writing about moderation (including some high-profile ban announcements), and as the amount of new content being posted to the site has more than tripled, much of it violating our AI content policy, the workload to enforce those policies has gone up drastically.

Backend refactor

The LessWrong codebase has for a while been buckling under its own weight. This wasn't a huge problem, but we noticed that some of the technical debt posed particularly big challenges for coding assistants trying to navigate the codebase. Also, it meant certain targets I wanted to hit for performance were very hard to achieve. 

So we fixed it! We despaghettified much of our codebase, and moved LessWrong to use the same framework as all of our other coding projects (Next.JS) and now performance is faster than it's ever been before, and coding agents seem to have much less trouble navigating our codebase, even accounting for how much smarter they've become over the course of the year. That said, this choice was controversial among my staff, so may history tell whether this was worth it. It did cost us like 8 staff months, which was pretty expensive as things go.

Overhauling our moderation tools

If you want to get a snapshot of the experience of being an internet moderator in 2025, I encourage you to go to lesswrong.com/moderation and check out the kind of content we get to sift through every day:

While I am glad that the language models seem to direct hundreds to thousands of people our way every day as one of the best places to talk about AI, they unfortunately are also assisting in sending huge mountains of slop that we have to sift through to make sure we don't miss out on the real humans who might end up great contributors to things we care about. 

To make processing the volume of submissions feasible, we revamped our moderation UI, and iterated on automated AI-detection systems to reduce our workload: 

The UI is modeled after Superhuman mail and has cut down our moderation processing time by something like 40%. 

The LessWrong frontpage feed

As the volume of content on LessWrong increases, it stops being in the interest of approximately anyone to simply see an ordered list of all recent comments on the frontpage of the site. I spend ~2 hours a day reading LessWrong and even I felt like the recent discussion section was failing to show me what I wanted to see!

And so we've been working on a feed for the frontpage of LessWrong that somehow tries to figure out what you want to read, selecting things to show you from both the archives of LessWrong and the torrent of content that gets posted here every day.

I am not totally satisfied with where the feed is currently at. I think the current algorithm is kind of janky and ends up showing me too many old posts, and too little of the recent discussion on the site. It also turns out that it’s really not clear how to display deeply nested comment trees in a feed, especially if you already read various parts of the tree, and I still want to iterate more on how we solve those problems. 

But overall, it’s clear to me we needed to make this jump at some point, and what we have now, despite its imperfections is a lot better than what we had previously.

Many UI improvements and bugfixes

And then we also just fixed and improved many small things all over LessWrong and fixed many many bugs. I really can't list them all here, but feel free to dig through the Github repository if you are curious.

Lighthaven

Now, the other big strand of our efforts is Lighthaven, our 30,000 sq. ft. campus in downtown Berkeley. 

It's really hard to estimate the positive externalities from a space like Lighthaven. However, I do really think it is helping a lot of the people who will have a big effect on how humanity navigates this coming decade have better conversations, have saner beliefs, and have better allies than they would have otherwise had. When I think of the 1000 most important conversations that happened in the Bay Area in the last year, my guess is a solid 1% of those happened at Lighthaven during one of our events.

The people love Lighthaven. It is a place of wonder and beauty and discourse—and I think there is a decent chance that no other physical place in the world will have ended up contributing more to humanity's grappling with this weird and crazy future we are now facing. 

And look, I want to be honest. Lighthaven was a truly crazy and weird and ambitious project to embark on. In as much as you are trying to evaluate whether Lightcone will do valuable and interesting things in the future, I think Lighthaven should be a huge positive update.

2025 has been the venue's busiest year to date. Here is a collage of many of the events we have hosted this year: 

Here is a calendar view of our publicly viewable events (we also had a small number of private events): 

Almost all of these Lighthaven events were put on by outside organizations, with Lightcone providing the venue and some ops support. However, Lightcone directly organized two events in 2025: LessOnline and Inkhaven.

LessOnline

We run an annual conference for the blogosphere named LessOnline. This year, over 600 people gathered together to celebrate being wrong on the internet. Many rationalist writers like Eliezer Yudkowsky, Scott Alexander, and Zvi Mowshowitz attend. Many rat-adjacent writers also love to come, including Patrick McKenzie, David Chapman, Michael Nielsen, Scott Sumner, and more. (...actually, the event caused Scott Aaronson to decide he's a rationalist now.)

People seem to be having a good time, leading activities/sessions, and meeting the blogosphere in person. Of the 360 people who filled out the feedback form, the average rating was 8.5, the NPS was 37, and the average rating for Lighthaven as a venue was 9.1—a good sign about the venue capacity, given that it was our biggest historical event!

The first year we ran it (2024), we made some basic mistakes and only just broke even on it. This year, we made a modest profit ($50k-$100k), and it also takes about half as much input from the core Lightcone team to organize, so we're in a good position to keep running this annually.

Inkhaven Residency

For the month of November, the internet came to Lighthaven. 41 great writers paid us to come gather at Lighthaven to write a blog post every day. The participants loved it, and even the Lightcone staff ended up publishing a bunch of IMO pretty great posts!

The program wasn't for everyone, but ratings are mostly high across the board, and the large majority would like to return when the opportunity presents itself. Another neat feature of a daily writing program is that a really huge number of participants wrote reflections on their experience, which really gives you a lot of surface area on how the program went: 

Inkhaven was the first program that really benefited from the fact that we run both lots of online publishing infrastructure, and a physical space like Lighthaven. The event ended up being both good for the world, and IMO also good for LessWrong. 

Of course, by its nature of being a program where you are supposed to publish daily, not all of those posts were of the same quality as other LessWrong posts. However, my guess is their average quality was actually substantially higher than the average quality of non-Inkhaven posts. This is evidenced by their average karma, as well as a sampling of November's top 20 posts by karma^[17]. 

Around 212 of the 748 posts in November 2026 were written by Inkhaven authors (almost 30%)! Those posts received an average of 50 karma, compared to 27.8 karma of non-Inkhaven posts. Also of the top 20 posts published by karma in November, 11 of them are Inkhaven posts (suggesting Inkhaven posts are ~2x as likely to end up in the top posts of the month compared to non-Inkhaven posts).

I also liked Garret Baker's thoughts on the effect of Inkhaven on LessWrong:

Garret's comment on Inkhaven and LessWrong

[...] I think LessWrong has become a much more vibrant and active place since Inkhaven started. Recently the frontpage has felt more... I can't think of a better word than "corporate"... than I'd like. Maybe what I mean is that the LessWrong posters have started catering more and more toward the lowest-common-LessWrong-denominator.

For example, here are the top posts from September 2025 (I think October had a reasonable amount of Inkhaven spirit, considering all the people doing Halfhaven)

Tomas is always nice and refreshing, but imo the rest of this is just really extremely uninteresting and uninspired (no offense to anyone involved, each post is on its own good I think, but collectively they're not that interesting), and seem very much catering to the lowest-common-LessWrong-denominator.

Contrast this selection with the following

I have not read as many of these (because there were more overall posts and these are more recent), but collectively the range of topics is so much more interesting & broad, you still get some lowest-common-denominator catering, but collectively these posts are so much more inspired than they were just two months ago. 

I really would very much enjoy a world where each person participating in Inkhaven continued posting every day even after Inkhaven.

Edit: I would also say that concretely this has changed how I use the website a whole bunch too. In September I think most of the interesting parts of LessWrong discussion was happening inside shortforms, but now I think most of the interesting discussion has moved to posts again, which I consider very healthy.

Inkhaven was very close to being economically self-sufficient^[18]. Participants seemed generally happy with the price, and the program was a decently good fit for sponsorship, which allowed us to get quite close to breaking even with the labor and Lighthaven costs of the program.

Finances

And that brings us to our finances! If you are thinking about giving us money, you'd want to know where we are spending money and what we would do with additional funds. So let's get down to brass tacks.

Some key highlights from 2025:

• Lighthaven got pretty close to breaking even!
• Most of our expenses are our core staff salaries
• With additional funds from SFF and the EA Infrastructure Fund in the middle of this year, we were able to make our deferred 2024 interest payment and did not have to take out a loan to do that^[19]

If anyone has any more detailed questions about anything, including our salary policies^[20], please feel free to ask in the comments!  

Lightcone and the funding ecosystem

An important (though often tricky to think about) question for deciding whether to donate to Lightcone is how that plays into other people funding us. 

Much of what I said last year about the funding ecosystem is still true. There don't really exist large philanthropic foundations that can fund our work besides SFF, and SFF is funding us (though only partially). Open Philanthropy (now Coefficient Giving) appears to continue to have some kind of blacklist for recommending grants to us, FTX of course no longer exists, and there are currently no other large philanthropic institutions that fund the kind of work we are doing. Various grantmaking programs for AI safety research have been popping up (such as the UK's ARIA program, or grant programs by capability labs), but they can only fund research directly, not any kind of research infrastructure.

Eli Rose, who was in charge of evaluating us at Open Phil/Coefficient Giving when they were still able to grant to us, said this last year on our fundraiser: 

I work as a grantmaker on the Global Catastrophic Risks Capacity-Building team at Open Philanthropy; a large part of our funding portfolio is aimed at increasing the human capital and knowledge base directed at AI safety. I previously worked on several of Open Phil’s grants to Lightcone.

As part of my team’s work, we spend a good deal of effort forming views about which interventions have or have not been important historically for the goals described in my first paragraph. I think LessWrong and the Alignment Forum have been strongly positive for these goals historically, and think they’ll likely continue to be at least into the medium term.

You can read the rest of his comment in this collapsible section:

All of this continues to put Lightcone in a position that is uniquely dependent on non-institutional donors to fill our funding gap. 

This situation honestly feels pretty crazy to me. Given how much of the broader ecosystem's funding flows through institutional funders, it seems kind of insane to have infrastructure as central as LessWrong and Lighthaven, and projects as transparently successful as AI 2027, get primarily funded by non-institutional donors, especially given that staff at places like Coefficient Giving and SFF and FLI think we are highly cost-effective. But as far as I know this is how it will be at least for 2026 (though we will certainly apply to whatever institutional funding programs are open to organizations like us, in as much as they exist)^[21].

Lightcone's work on improving the funding landscape

Historically one of our big strains of work has been our efforts on improving the philanthropic funding landscape around AI existential risk and adjacent areas. I summarized our work in the space like this in last year's fundraiser: 

The third big branch of historical Lightcone efforts has been to build the S-Process, a funding allocation mechanism used by SFF, FLI and Lightspeed Grants.

Together with the SFF, we built an app and set of algorithms that allows for coordinating a large number of independent grant evaluators and funders much more efficiently than anything I've seen before, and it has successfully been used to distribute over $100M in donations over the last 5 years. Internally I feel confident that we substantially increased the cost-effectiveness of how that funding was allocated—my best guess is on the order of doubling it, but more confidently by at least 20-30%^[18], which I think alone is a huge amount of good done.^[19]

Earlier this year, we also ran our own funding round owned end-to-end under the banner of "Lightspeed Grants": 

Somewhat ironically, the biggest bottleneck to us working on funding infrastructure has been funding for ourselves. Working on infrastructure that funds ourselves seems ripe with potential concerns about corruption and bad incentives, and so I have not felt comfortable applying for funding from a program like Lightspeed Grants ourselves. Our non-SFF funders historically were also less enthusiastic about us working on funding infrastructure for the broader ecosystem than our other projects.

We did not do as much work this year on improving the funding landscape as we have done in previous years, largely owing to lack of funding for development of that kind of infrastructure! 

I have however continued to be an active fund manager on the Long Term Future Fund, which has continued to make a substantial number of high-quality grants over the course of the year, and we have picked up one important strain of work that I am hoping to develop more in 2026.

The AI Risk Mitigation Fund (ARM Fund) was founded in 2024 by members of the Long Term Future Fund, as a fund more specifically dedicated to reducing catastrophic risk from AI. This year the ARM Fund moved from being a project fiscally sponsored by Emergent Ventures to becoming a program of Lightcone Infrastructure (but still under substantial leadership from Caleb Parikh). 

Our exact plans with the ARM Fund are up in the air, but you can read more about them in our "Plans for 2026" section.

Lightcone as a fiscal sponsor

Another thing we have started doing this year is to help a few organizations and projects that we think are particularly valuable by providing them with fiscal sponsorship as part of our 501(c)3 status, and access to some of our internal infrastructure for accounting and financial automations. 

We largely did this because it seemed like an easy win, and it's not a part we are looking to scale up, but I figured I would still mention it. We are currently providing fiscal sponsorship for: 

• Adam Scholl and Aysja Johnson's research
• John Wentworth and David Lorell's research
• AI Lab Watch
• Charitable research by Elizabeth Van Nostrand
• Rationality Meetup organizing by Skyler Crossman
• LessWrong Deutschland for meetup organizing
• FlexHeg builders

Fiscal sponsors usually take between 6% and 12% of revenue as a sponsorship fee. We don't currently charge fees and are probably a better fiscal sponsor than alternatives. We have received around $1M in donations for the above projects in 2025, so if you think the above projects are cost-effective, a reasonable estimate is that we generated something like $60k to $120k in value via doing so.

If it's worth doing, it's worth doing with made up statistics

Thus is it written: “It’s easy to lie with statistics, but it’s easier to lie without them.”

Last year, I did a number of rough back of the envelope estimates to estimate how much impact we produced with our work. Their basic logic still seems valid to me and you can read them in the collapsible section below, but I will also add some new ones for work we did in 2025. 

Last year's back of the envelope estimates

Ok, so I've waffled about with a bunch of high-level gobbledigook, but as spreadsheet altruists the only arguments we are legally allowed to act on must involve the multiplication of at least 3 quantities and at least two google spreadsheets.

So here is the section where I make some terrible quantitative estimates which will fail to model 95% of the complexity of the consequences of any of our actions, but which I have found useful in thinking about our impact, and which you will maybe find useful too, and which you can use to defend your innocence when the local cost-effectiveness police demands your receipts.

The OP GCR capacity building team survey

Open Philanthropy has run two surveys in the last few years in which they asked people they thought were now doing good work on OP priority areas like AI safety what interventions, organizations and individuals were particularly important for people getting involved, or helped people to be more productive and effective. 

Using that survey, and weighting respondents by how impactful Open Phil thought their work was going to be, they arrived at cost-effectiveness estimates for various organizations (to be clear, this is only one of many inputs in OPs grantmaking). 

In their first 2020 survey, here is the table they produced^[22]: 

Org	$/net weighted impact points (approx; lower is better)
SPARC	$9
LessWrong 2.0	$46
80,000 Hours	$88
CEA + EA Forum	$223
CFAR	$273

As you can see, LessWrong 2.0's impact was in estimated cost-effectiveness only behind SPARC (which is a mostly volunteer driven program, and this estimate does not take into account opportunity cost of labor).

In their more recent 2023 survey, Lightcone's work performed similarly well. While the data they shared didn't include any specific cost-effectiveness estimates, they did include absolute estimates on the number of times that various interventions showed up in their data: 

These are the results from the section where we asked about a ton of items one by one and by name, then asked for the respondent’s top 4 out of those. I’ve included all items that were listed more than 5 times.

These are rounded to the nearest multiple of 5 to avoid false precision.

80,000 Hours	125
University groups	90
EAGs/EAGxes	70
Open Philanthropy	60
Eliezer's writing	45
LessWrong (non-Eliezer)	40
[...]
Lightcone (non-LW)	15

To get some extremely rough cost-effectiveness numbers out of this, we can divide the numbers here by the budget for the associated organizations, though to be clear, this is definitely an abuse of numbers.

Starting from the top, during the time the survey covered (2020 - early 2023) the annual budget of 80,000 Hours averaged ~$6M. Lightcone's spending (excluding Lighthaven construction, which can't have been relevant by then) averaged around $2.3M. University groups seem to have been funded at around $5M/yr^[23], and my best guess is that EAG events cost around $6M a year during that time. I am going to skip Open Philanthropy because that seems like an artifact of the survey, and Eliezer, because I don't know how to estimate a reasonable number for him.

This produces this table (which I will again reiterate is a weird thing to do): 

Project	Mention	Mentions / $M
80,000 Hours	125	6.4
University groups	90	5.
EAGs/EAGxes	70	3.6
Lightcone (incl. LW)	40 + 15	6.8

As you can see, my totally objective table says that we are the most cost-effective intervention that you can fund out there (to be clear, I think the central takeaway here is more "by this very narrow methodology Lightcone is competitive with the best interventions, I think the case for it being the very best is kind of unstable")

Lightcone/LessWrong cannot be funded by just running ads

An IMO reasonable question to ask is "could we fund LessWrong if we just ran ads?" It's not fully clear how that relates to our cost-effectiveness, but I still find it a useful number to look at as a kind of lower-bound on the value that LessWrong could produce, with a small change.

LessWrong gets around 20 million views a year, for around 3 million unique users and 12 million engagement minutes. For our audience (mostly American and English-speaking) using Google AdSense, you would make about $2 per 1000 views, resulting in a total ad revenue of around $40,000, a far cry from the >$1,000,000 that LessWrong spends a year. 

Using Youtube as another benchmark, Youtubers are paid about $15 for 1000 U.S. based ad impressions, with my best guess of ad frequency on Youtube being about once every 6 minutes, resulting in 2 million ad impressions and therefore about $30,000 in ad revenue (this is ignoring sponsorship revenue for Youtube videos which differ widely for different channels, but where my sense is they tend to roughly double or triple the default Youtube ad revenue, so a somewhat more realistic number here is $60,000 or $90,000).

Interestingly, this does imply that if you were willing to buy advertisements that just consisted of getting people in the LessWrong demographic to read LessWrong content, that would easily cover LessWrong's budget. A common cost per click for U.S. based ads is around $2, and it costs around $0.3 to get someone to watch a 30-second ad on Youtube, resulting in estimates of around $40,000,000 to $4,000,000 to get people to read/watch LessWrong content by just advertising for it.

Comparing LessWrong to other websites and apps

Another (bad) way of putting some extremely rough number of the value LessWrong provides to the people on it, is to compare it against revenue per active user numbers for other websites and social networks.

Platform	U.S. ARPU (USD)	Year	Source
Facebook	$226.93 (Annual)	2023	Statista
Twitter	$56.84 (Annual)	2022	Statista
Snapchat	$29.98 (Annual)	2020	Search Engine Land
Pinterest	$25.52 (Annual)	2023	Stock Dividend Screener
Reddit	$22.04 (Annual)	2023	Four Week MBA

I think by the standards of usual ARPU numbers, LessWrong has between 3,000 and 30,000 active users. So if we use Reddit as a benchmark this would suggest something like $75,000 - $750,000 per year in revenue, and if we use Facebook as a benchmark, this would suggest something like $600,000 - $6,000,000. 

Again, it's not enormously clear what exactly these numbers mean, but I still find them useful as very basic sanity-checks on whether we are just burning money in highly ineffectual ways.

Lighthaven event surplus

Over the last year, we negotiated pricing with many organizations that we have pre-existing relationships with using the following algorithm: 

1. Please estimate your maximum willingness to pay for hosting your event at Lighthaven (i.e. at what price would you be indifferent between Lighthaven and your next best option)
2. We will estimate the marginal cost to us of hosting your event
3. We use the difference between these as an estimate of the surplus produced by Lighthaven and we split it 50/50, i.e. you pay us halfway between our marginal cost and your maximum willingness to pay

This allows a natural estimate of the total surplus generated by Lighthaven, measured in donations to the organizations that have hosted events here.

On average, event organizers estimated total value generated at around 2x our marginal cost. 

Assuming this ratio also holds for all events organized at Lighthaven, which seems roughly right to me, we can estimate the total surplus generated by Lighthaven. Also, many organizers adjusted the value-add from Lighthaven upwards after the event, suggesting this is an underestimate of the value we created (and we expect to raise prices in future years to account for that).

This suggests that our total value generated this way is ~1.33x our revenue from Lighthaven, which is likely to be around $2.8M in the next 12 months. This suggests that as long as Lighthaven costs less than ~$3.72M, it should be worth funding if you thought it was worth funding the organizations that have hosted events and programs here (and that in some sense historical donations to Lighthaven operate at least at a ~1.33x multiplier compared to the average donation to organizations that host events here).

To help get a sense of what kind of organizations do host events here, here is an annotated calendar of all the events hosted here in 2024, and our (charitable) bookings for 2025: 

Cost-Effectiveness Estimates for 2025 work

It's tricky to evaluate our work on communicating to the public in terms of cost-effectiveness. It's easy to measure salience and reach, but it's much harder to measure whether the ideas we propagated made things better or worse.

So I will focus here on the former, making what seem to me relatively conservative assumptions about the latter, but assuming that the ideas we propagated ultimately will improve the world instead of making it worse, which I definitely do not think is obvious!

For LessWrong in particular, I think last year's estimates are still pretty applicable, and given the increase in traffic and activity, you can basically multiply the numbers for those estimates by roughly 1.5x, and since nothing else really changed, I won't repeat the analysis here.

AI 2027

AI 2027 focused on things that matter, and there are few things that I would consider more valuable for someone to read than AI 2027. It also made an enormous splash in the world.

Indeed, the traffic for AI 2027 in 2025 has been similar to the traffic for all of LessWrong in 2025, with about 5M unique users and 10M page-views, whereas LessWrong had 4.2M unique users and 22M page-views. 

Comparing to paid advertisements

A reference class that seems worth comparing against is how much it would have cost to drive this much traffic to AI 2027 using advertisement providers like Google AdSense. Ad-spend is an interesting comparison because in some sense driving attention to something via paid advertisements is the obvious way of converting money into attention, and how much of global attention gets allocated.

Cost per click tends to be around $2 per click for the demographics that AI 2027 is aimed at, so driving 10M pageviews to AI 2027 using classical online advertising would have cost you on the order of $20M, which of course would pretty easily fully fund both Lightcone's and the AI Futures team's budgets.

Another related reference class is how much money AI 2027 would have generated if we simply had put a Google AdSense ad on it. Historically much of journalism and the attention economy has been driven by advertisement revenue.

Google AdSense ads tend to generate around $2 per 1,000 views, so you would end up with around $20,000. If you fully count secondary media impressions as well (i.e. count views for AI 2027-based YouTube videos and news articles), you would end up at more like $200,000. This is not nothing, but it's a far cry from what the project actually cost (roughly $750k across Lightcone and the AI Futures team).

Comparison to think tank budgets

Consider a hypothetical think tank with a budget on the order of $100M. If the only thing of significance that they ended up doing in the last few years of operations was release AI 2027, I think funding them would have been a cost-effective donation opportunity.

For example, CSET was funded for around $100M by Open Philanthropy/Good Ventures. I would trade all public outputs from CSET in exchange for the reach and splash that AI 2027 made.

And I do think some of CSET's work has been quite impactful and valuable! Their work on the semiconductor supply chain did raise a lot of important considerations to policymakers. Of course, it's harder to quantify the other benefits of a program like CSET, which e.g. included allowing Jason Matheny to become head of RAND, which has an annual budget of around $500M (though of course only a very small fraction of that budget goes to projects I would consider even remote candidates for being highly cost-effective). But I still find it a valuable comparison. 

Comparison to 80,000 Hours

80,000 Hours has an annual budget of around $15M. The primary output of 80k is online content about effective career choice and cause prioritization. They use engagement-hours as the primary metric in their own cost-effectiveness estimates. Their latest impact report cites around 578,733 engagement hours. 

Meanwhile, AI-2027.com has had 745,423 engagement hours in 2025 so far, and by the end of the year will have likely produced 50% more engagement hours than all of 80k's content in 2024. 

80k's latest report goes through June 2025. A large portion of 80k's engagement hours in this report were from their AI 2027 video.

This suggests that 80k themselves considers engagement with the AI 2027 content to be competitive with engagement with their other material. If you were to straightforwardly extrapolate 80k's impact from their engagement hours, this would suggest a substantial fraction of 80k's total impact for 2025, as measured by their own metrics, will be downstream of AI 2027.

This naively suggests that if you think 80k was a cost-effective donation target, then it would make sense to fund the development and writing of AI 2027 at least to the tune of $15M (though again, this kind of estimate is at most a rough sanity-check on cost-effectiveness estimates and shouldn't be taken at face value). By my previous estimate of Lightcone being responsible for roughly 30% of the impact of the project, this would justify our full budget 2-3 times over.

Lighthaven in 2025

A conservative estimate of the value we provide is to assume that Lightcone is splitting the surplus generated by Lighthaven 50/50 with our clients, as that split was roughly what we were aiming for when we set our event pricing baselines in 2024 (see last year's estimates for more detail on that): 

We project ending 2025 with a total of ~$3.1M in event revenue. My algorithm for splitting the marginal benefit 50/50 with our clients suggests that as long as Lighthaven costs less than ~$4.12M (which it did), it should be worth funding if you thought it was worth funding the organizations that have hosted events and programs here. 

I estimate that Lighthaven will generate $100k–$200k more revenue in 2026 compared to in 2025.^[24]

What other people say about funding us

Independent evaluation efforts are rare in the AI safety space. People defer too much to fads, or popularity, or large institutions on which projects are worth funding. But ultimately, it is really hard to fully independently evaluate a project like Lightcone yourself, so here is a collection of things I could find that people have said about our cost-effectiveness.

If you know of any other recent public evaluations of our cost-effectiveness, post them in the comments! I tried to be thorough in listing everything I could find that takes any kind of stance on the value of donating to Lightcone.

The Lightcone Principles

A key consideration for deciding whether Lightcone is worth supporting, is to understand what kind of principles determine Lightcone's priorities and operational decisions. 

Conveniently, as part of Inkhaven, I have just published a sequence of internal memos that try to illustrate our core operational principles. The sequence is not finished yet, but the essays released so far are: 

• "Question the Requirements" 
• "Do not hand off what you cannot pick up" 
• "Two can keep a secret if one is dead. So please share everything with at least one person"
• "Tell people as early as possible it's not going to work out"
• "Increasing returns to effort are common"
• "Put numbers on stuff, all the time, otherwise scope insensitivity will eat you"
• "Close open loops"
• "Aim for single piece flow"
• "Automate, automate it all"
• "Be Naughty"

For another (somewhat more outdated) look into our operational philosophy, you can read this collection of short stories that Aaron Silverbook wrote about us: "The Adventures of the Lightcone Team".

Most importantly, we value flexibility and the ability to pivot extremely highly. This has historically meant that we grow very slowly and put a huge premium on wide generalist skillsets in the people we hire. 

This paid off a lot this year, where we (probably) ended up having a lot of our impact by working on things like AI 2027, the MIRI book, and Deciding To Win. It's the philosophy that allowed us to build Lighthaven even though we had largely been a team of web-developers before, and build out much of the S-Process and the Survival and Flourishing Fund the years before that.

Plans for 2026

What we will do next year will depend a lot on how much funding we can raise.

The very minimum with which we can survive at all is roughly $1.4M. If we raise that much but not more, we will figure out how to take out a mortgage on the one property we own next to Lighthaven. This will increase our annual burn rate by around $70k/year in interest payments, will cost us a few staff weeks to get arranged, and will add a bunch of constraints to how we can use that property which will probably reduce Lighthaven income by another $30k–$50k a year in expectation. 

If we fundraise less than $1.4M (or at least fail to get reasonably high confidence commitments that we will get more money before we run out of funds), I expect we will shut down. We will start the process of selling Lighthaven. I will make sure that LessWrong.com content somehow ends up in a fine place, but I don't expect I would be up for running it on a much smaller budget in the long run.

$2M is roughly the "we will continue business as usual and will probably do similarly cool things to what we've done in the past" threshold. In addition to continuing all of our existing development of LessWrong, Lighthaven, and collaborations with the AI Futures team and MIRI, this amount will likely include at least one additional major project and 1–2 additional hires to handle that increased workload.

I do have a lot of things I would like to invest in. While I don't expect to make a lot of core hires to our organization, I would like to trial a bunch of people. I see a lot of ways we could substantially increase our impact by spending more money and paying various contractors to expand our range of motion, and of course given that at this present moment we have less than a month of runway in the bank, it would really be very valuable for us to have more runway so we don't immediately drop dead in exactly 12 months.

Here are some projects that are at the top of my list of things we are likely to work on with more funds: 

LessWrong LLM tools

We did an exploratory ~2 month sprint in late 2024 on figuring out how to use language models as first class tools on LessWrong (and for intellectual progress more broadly). While we found a few cool applications (I continue to use the "base model" autocomplete tool that we built for admins in the LessWrong editor regularly whenever I write a long post in order to get me unstuck on what to say next), it seemed to me that making something actually great with the models at the time would take a huge amount of work that was likely going to be obsolete within a year. 

But it's been more than a year, and the models are really goddamn smart now, and I can think of very few people who are better placed to invest in LLM-thinking tools than we are. The LessWrong community is uniquely open to getting value out of modern AI systems, we have a lot of early-adopter types who are willing to learn new tools, and we have almost a decade of experience building tools for collective thought here on LessWrong. 

Figure out how to deploy money at scale to make things better

I currently think it's around 35% likely that Anthropic will IPO in 2026. Even if it doesn't, I expect Anthropic will have some kind of major buy-back event in 2026. 

If an IPO happens at Anthropic's ~$300B valuation, together with Anthropic's charitable 3:1 matching program for donations from early employees, this seems like it would result in by far the biggest philanthropic scale-up in history. It would not surprise me if Anthropic employees will collectively be trying to distribute over $5B/yr by end of 2026, and over $20B/yr by end of 2027. This alone would end up being on the order of ~5% of current total philanthropic spending in the US^[26]. 

OpenAI employees will likely be less aggressive in their philanthropic spending, but my guess is we will also see an enormous scale up of capital deployed by people who have at least some substantial interest in reducing existential risk and shaping the long-run future of humanity. 

Lightcone has been working on grant-making infrastructure for 7+ years, which puts us in a good position to build something here. We also have a relatively adversarial stance toward the labs, making us not a great fit to distribute all of this capability labs money. Nevertheless, I am not seeing anyone else picking up the ball here who I expect to do a good job.

As part of our early efforts in the space, the ARM Fund (airiskfund.com) has recently become a Lightcone sponsored project, and we will be investing staff time to scale it up and make it better. In particular, I am excited about creating infrastructure for many funders to coordinate their giving, negotiate fair funding shares for different projects, and create consistent deal-flow where promising projects can be put in front of the eyes of potential funders, and any advisors they want to defer to. The S-Process infrastructure we built between 2019 and 2024 has taught us a lot of valuable lessons about what works and doesn't work in the space.

More and more ambitious public communication about AI Existential risk

We will be working with the AI Futures team on the sequel to AI 2027, but there are many more opportunities for high quality public communications about ASI strategy and risk.

For example, it's pretty plausible we will want to expand directly into video production. IMO the 80k YouTube team has been a big success, and there is a lot more demand for producing high quality video content on AI existential risk and related topics. 

And even within the domain of internet essay production, I think there are many more high-quality beige websites that should exist on a wide variety of topics, and I think the returns in terms of shifting public opinion and information key decision-makers continue to be very high.

What do you get for donating to Lightcone Infrastructure?

The most important thing you get for donating to Lightcone are the fruits of our labor. A world (hopefully) slightly less likely to go off the rails. More smart people thinking openly and honestly about difficult topics. A civilization more on the ball of what matters in the world. 

But, that's not the only thing! We try pretty hard to honor and credit the people whose financial contributions have made it possible for us to exist.

We do this in a few different ways: 

• If you donate more than $1,000 to us, we will send you a limited edition Lightcone donor t-shirt (if you fill out the form we send you). You can also claim an entry on the LessWrong donor leaderboard.
• If you donate more than $2,000, you can get a personalized plaque on our campus.
• If you donate more than $20,000, you can get a structure on campus, like one of our meeting rooms, a gazebo, or some kind of room named after you.
• For even larger donations, you could have a whole lecture hall or area of campus named after you (like the Drethelin Gardens, dedicated to Drethelin as a result of a $150k donation to us).

And if you want any other kind of recognition, please reach out! It's pretty likely we can make something work.

Logistics of donating to Lightcone

We are a registered 501(c)3 nonprofit. If you are in the United States, the best way to donate cash, stock, or crypto is through every.org:

https://www.every.org/lightcone-infrastructure/f/2026-fundraiser 

We are also set up to readily receive tax-deductible donations from Germany, Switzerland, and the United Kingdom. 

If you are in Germany or Switzerland, donate via Effektiv-Spenden.de. Send an email to me (habryka@lightconeinfrastructure.com) and we will provide the info you need to send the funds and get that sweet sweet government tax reduction.

If you are in the UK, donate through NPT Transatlantic to get UK tax deductibility. 

If there is enough interest, I can probably also set up equivalence determinations in most other countries that have a similar concept of tax-deductibility. 

If you have any weirder assets that you can't donate through every.org, we can figure out some way of giving you a good donation receipt for that. Just reach out (via email, DM, or text/signal at +1 510 944 3235) and I will get back to you ASAP with the logistics. Last round we received a bunch of private company equity that I expect we will sell this year and benefit quite a bit from, while providing some healthy tax-deductibility for the donor.

Also, please check if your employer has a donation matching program! Many big companies double the donations made by their employees to nonprofits (for example, if you work at Google and donate to us, Google will match your donation up to $10k). Here is a quick list of organizations with matching programs I found, but I am sure there are many more. If you donate through one of these, please also send me an email so I can properly thank you and trace the source of the donation.

Every.org takes 1% of crypto donations through the platform, so if you want to donate a lot of crypto, or don't want your identity revealed to every.org, reach out to us and we can arrange a transfer.

Thank you

And that's it. If you are reading this, you have read (or at least scrolled past) more than 13,000 words full of detail about what we've done in the last year, what our deal is, and what will happen if you give us money. 

I do really think that we are very likely the best option you have for turning money into better futures for humanity, and hopefully this post will have provided some evidence to help you evaluate that claim yourself. 

I think we can pull the funds together to make sure that Lightcone Infrastructure gets to continue to exist. If you can and want to be a part of that, donate to us here. We need to raise ~$1.4M to survive the next 12 months, $2M to continue our operations as we have, and can productively use a lot of funding beyond that. 

1. ^{^}

   We built the website and branding, the AI Futures Team wrote the content and provided the data

2. ^{^}

   The Survival and Flourishing Fund continues to recommend grants to us in a way that indicates their process rates us as highly cost-effective, and staff responsible for evaluating us at Coefficient Giving (previously Open Philanthropy) think it’s likely we continue to be highly cost-effective (but are unable to recommend grants to us due to political/reputational constraints of their funders). The EA Infrastructure Fund also recommended a grant to us, though is unlikely to do so going forward as their leadership is changing and is being integrated into the Centre for Effective Altruism.

3. ^{^}

   Here are some individuals who have personally given to us or strongly recommended donations to us who others might consider well-informed about funding opportunities in this space: Zvi Mowshovitz, Scott Alexander, Ryan Greenblatt, Buck Shlegeris, Nate Soares, Eliezer Yudkowsky, Vitalik Buterin, Jed McCaleb, Emmett Shear, Richard Ngo, Daniel Kokotajlo, Steven Byrnes and William Saunders. 

4. ^{^}

   Part of that is around $300k of property tax, which we are currently applying for an exemption for. Property tax exemption applications seem to take around 2-3 years to get processed (and we filed ours a bit more than a year ago). If we get granted an exemption we will get any property tax we paid while the application was pending refunded, so if that happens the project will have basically broken fully even this year!

5. ^{^}

   See last year's fundraiser post for more detail on our finances here.

6. ^{^}

   The fundraising progress thermometer will include these matching funds. I also go into why it's 12.5%, which is a weirdly specific number, in the finances section later in the post. 

7. ^{^}

   To be clear, I do not expect this to continue to be the case for the coming decade. As the financial stakes grow larger, I expect discussion of these topics to start being dominated by ideas and memes that are more memetically viable. More optimized for fitness in the marketplace of ideas.

8. ^{^}

   It gives a model of the AI Alignment problem that I think is hilariously over-optimistic. In the "Slowdown" scenario, humanity appears to suddenly end up with aligned Superintelligence in what reads to me as an extremely unrealistic way, and some of the forecasting work behind it is IMO based on confused semi-formal models that have become too complicated for even the authors to comprehend.

9. ^{^}

   And to be clear, Lightcone did receive some payment for our work on AI 2027 from the AI Futures Team (around $100k). The rate we offered was not a sustainable rate for us and the time we spent cost us more in salaries than that. This low rate was the result of the AI Futures team at the time also being quite low on funding. The AI Futures team also has some of their own credit allocation estimates which put Lightcone closer to 15% instead of 30%. 

10. ^{^}

    And here Lightcone did receive a pretty substantial amount of money for our work on the project (around $200k).

11. ^{^}

    Also legally we have some constraints on what we can work on as a 501(c)3, which generally cannot engage in lobbying for specific candidates. We were in the clear for this project since we didn't work on any of the content of the website, and were paid market rate for our work (which is higher than our rates for AI 2027 or If Anyone Builds It), but we would have entered dicey territory if we had been involved with the content.

12. ^{^}

    Which to be clear, I can't really comprehensively argue for in this post

13. ^{^}

    Unfortunately due to the rise of AI agents, Google Analytics has become a much worse judge of traffic statistics than it has been in previous years, so I think these specific numbers here are likely overestimating the increase by approximately a factor of 2.

14. ^{^}

    As a quick rundown: Shane Legg is a Deepmind cofounder and early LessWrong poster directly crediting Eliezer for working on AGI. Demis has also frequently referenced LW ideas and presented at both FHI and the Singularity Summit. OpenAI's founding team and early employees were heavily influenced by LW ideas (and Ilya was at my CFAR workshop in 2015). Elon Musk has clearly read a bunch of LessWrong, and was strongly influenced by Superintelligence which itself was heavily influenced by LW. A substantial fraction of Anthropic's leadership team actively read and/or write on LessWrong.

15. ^{^}

    For a year or two I maintained a simulated investment portfolio at investopedia.com/simulator/ with the primary investment thesis "whenever a LessWrong comment with investment advice gets over 40 karma, act on it". I made 80% returns over the first year (half of which was buying early shorts in the company "Nikola" after a user posted a critique of them on the site). 

    After loading up half of my portfolio on some option calls with expiration dates a few months into the future, I then forgot about it, only to come back to see all my options contracts expired and value-less, despite the sell-price at the expiration date being up 60%, wiping out most of my portfolio. This has taught me both that LW is amazing alpha for financial investment, and that I am not competent enough to invest on it (luckily other people have done reasonable things based on things said on LW and do now have a lot of money, so that's nice, and maybe they could even donate some back to us!)

16. ^{^}

    Off-the-cuff top LW worldview predictions: AI big, prediction markets big, online communities big, academia much less important, crypto big, AI alignment is a real problem, politics is indeed the mindkiller, god is indeed not real, intelligence enhancement really important, COVID big

    Off-the-cuff worst LW worldview predictions: Politics unimportant/can be ignored, neural nets not promising, we should solve alignment instead of trying to avoid building ASI for a long while, predicted COVID would spread much more uncontrollably than it did

17. ^{^}

    In general, I am pretty sure LessWrong can handle many more good-faith submissions in terms of content right now, and trust the karma system to select the ones worth reading. HackerNews handles almost two orders of magnitude more submissions than we have in a day, and ends up handling it fine using the same fundamental algorithm.

18. ^{^}

    The cost of Inkhaven is probably dominated by the opportunity cost of lost clients during the month of November. We managed to run one major unrelated conference on one weekend, but could've probably found at least a few other smaller events to make money from. Inkhaven itself brought $2k-$4k per resident, though around half of them required financial assistance; it also brought in ~$75k in sponsorships, and of course we had to pay a lot for staff fees. On net I think Inkhaven cost us around $50k primarily due to the opportunity cost of not renting out Lighthaven to someone else.

19. ^{^}

    We deferred one of our annual $1M interest payments on Lighthaven from 2023 to 2024, since we were dealing with the big FTX-caused disruption to the whole funding ecosystem. This meant we were trying to fundraise for $3M last year to make both of our interest payments, but we only raised just above $2M by the end of our fundraiser. 

    Our backup plan was to take out a loan secured against a property we own right next to Lighthaven, which is worth around $1M. This would have been quite costly to us because commercial lenders really don't like loaning to nonprofits and the rates we were being offered were pretty bad. I am glad we didn't have to do it!

20. ^{^}

    Our salary policy is roughly that people at Lightcone are paid approximately 70% of what I am quite confident they could make if they worked in industry instead of at Lightcone, though we have not increased salaries for quite a while and we have been drifting downwards from that. See here for last year's discussion of our salary policies. 

21. ^{^}

    And we did receive a grant from the EA Infrastructure Fund which I am really grateful for. I do think given upcoming change in leadership of the fund this is unlikely to repeat (as I expect the new leadership to be much more averse to reputational associations to us).

22. ^{^}

    This table is not exhaustive and OpenPhil told us they chose organisations for inclusion partially dependent on who it happened to be easy to get budget data on. Also, we've removed one organization at their request (which also ranked worse than LessWrong 2.0).

23. ^{^}

    The linked grant is ~6M over a bit more than 2 years, and there are a bunch of other grants that seem also to university groups, making my best guess around $5M/yr, but I might be off here.

24. ^{^}

    Up until recently I was expecting a larger revenue increase in 2026 compared to 2025, but I now think this is less likely. The MATS program, which made up around $1.2M of our revenue while booking us for a total of 5 months for 2025, is moving to their own space for their 2025 summer program. This means we are dealing with a particularly large disruption to our venue availability. 

    While MATS makes up a lot of our annual revenue, the program's revenue per day tends to be lower than almost every other program we host. I expect that, in the long run, MATS leaving will increase the total revenue of the venue, but we likely won't be able to fully execute on that shift within 2026.

25. ^{^}

    Ryan later donated an extra $30k on top of the $3,000 described in this comment.

26. ^{^}

    I expect this willingness to spend to probably translate into only about 50% of this money actually going out, mostly because spinning up programs that will be able to distribute this much money will take a while, and even remotely-reasonable giving opportunities will dry up. 

27. ^{^}

    By "probably" I do just mean something like "a bit above 50%", I am very much not confident in this, but it's my best guess!

28. ^{^}

    https://ifanyonebuildsit.com/

29. ^{^}

    https://www.cold-takes.com/most-important-century/

30. ^{^}

    https://ai-2027.com/