Running Lightcone Infrastructure, which runs LessWrong and Lighthaven.space. You can reach me at habryka@lesswrong.com.
(I have signed no contracts or agreements whose existence I cannot mention, which I am mentioning here as a canary)
habryka
Karma: 52,546
I agree it is technically an evaluation of our cost-effectiveness! Though not much more than generic insults are, which I wouldn’t feel much of a need to link. I think it was reasonable for you to link it here, and also don’t think it makes sense to include it in the post.
Running Less Wrong doesn’t (have to) cost 1.4M per year. It’s a website. Websites of this scale can be run at 1/10th the cost, or even 1/100th the cost if you’re frugal. Having extremely onerous community management practices is a choice you’re making.
This is just really not true. LessWrong is a software project of substantial complexity, with a substantially large user-base. I don’t know any other platform that does anything remotely similar to what LessWrong does and supposedly would run at anything like 1/100th or 1/10th the cost.
Of course, if you just grab a forum software off-the-shelf and don’t bother moderating then you can run things much cheaper, but this is a strategy that generally does not work, and does not produce anything with the cultural or social impact of something like LessWrong. Online forums have been declining across the board for decades. There exist very few alive and growing internet forums. It is an exceptionally difficult task to make a thing like LessWrong grow and to continue producing a huge amount of research and intellectual contributions when your competitors are hyper-optimized social media platforms.
The problem that LessWrong is trying to solve does not have an off-the-shelf solution. Of course it would be great if it had, and in that case the high cost would be born by whoever developed that off-the-shelf solution, but there really isn’t one.
It’s a website.
I mean, Facebook is a website. Google is a website. ChatGPT is a website. Airtable is a website. Figma is a website. I don’t really know what this means. More than 10% of U.S. GDP is spent on “websites”. As a category they are not cheap to run, and the software behind them is not easy or cheap to build. I am not saying there are no valid critiques to make here, but the reference class of “website” is certainly not particularly helpful here.
For some more discussion on these topics see last year’s fundraiser post comments:
CowardlyPersonUsingPseudonym proposing we cut salaries in half
David Matolsci discussing the funding breakdown of Lightcone
If someone believes that a large part of your value is in hosting LW, and what you do outside this pales in importance (or may even be negative!), then there’s a bundling issue for that person. It’s impossible to donate to the upkeep of LW without donating to all that other stuff, correct?
Yep, it’s a bundling problem, though a pretty common one. You can’t invest in Deepmind, you can only buy Google stock. In general organizations benefit a lot from the ability to re-allocate resources between different projects. I certainly don’t want to commit to not investing in new projects.
Like, from a decision-making perspective it’s not even clear how accepting funds just for the purpose of donating to LW would work. Money is fungible, and as long as we receive fewer funds earmarked such than what we are planning to spend on LW anyways, this wouldn’t really affect our decisions at all, and I think it’s very unlikely we would end up receiving more funds than that.
I do care non-trivially about what our donors think and so saying you would like us to spend more resources on LW is a good thing to say at the same time as you donate!
Makes sense! Options like giving what we can regranting seem to work for a bunch of people but maybe they aren’t willing to do it for lightcone?
GWWC wasn’t accepting new applications for organizations to sponsor this way last time I checked a few months ago. I asked them to let me know when they do, and haven’t heard back, but I should ping them again and double check!
Yep, agree that it’s better to be funded by a collection of 20+ disagreeable frontier lab employees than 100% Open Phil, but I do also pretty intensely regret relying so much on Open Phil funding, so this is compatible with my state of belief.
I think social group beliefs are generally very correlated with each other, and think it will likely be the case that if we rely very heavily on frontier lab employee funding that there will be a lot of implicit pressure that will be backed by a threat of successful coordination to not fund us if we disagree.
(Also, to give a bit more context on my thinking here, I currently think that it’s fine for us to accept funding from Deepmind safety employees without counting towards this bucket, largely because my sense is the social coordination across the pond here is much less intense, and generally the Deepmind safety team has struck me as the most independent from the harsh financial incentives here to date)
Ah, this is a great point. I will update the text of the article with some guidance to this effect.
Agree it would be good to set up some better way for small to medium UK based donors to donate tax deductibly, but haven’t found a good way of doing so that wouldn’t cost me 40+ hours.
That… I think is below my bar for having any actual logical content instead of just snark, but if a few people want to second that they would have somehow found this valuable to read, I can add it to the post.
how to set up some kind of good quality AI Safety Forum that Coefficient is willing to fund.
There are millions of non-OP dollars in the ecosystem. Even without capability money, trying to start a new forum because of some political preferences by Dustin seems both dumb on a tactical level, and dumb on an incentive level (the last thing we want is for infrastructure projects to be downstream of random US political drama).
how likely do you think it is that after the IPO a few rich Anthropic employees will just cover most of Lightcone’s funding need?
I don’t want Lightcone to be >50%[1] funded by frontier lab employees (with some narrow exceptions), so that will dampen effects from things like an Anthropic IPO. I do still think if there is a lot of Anthropic money around this will free up other money, which will still make fundraising for us easier.
- ^
I am not yet fully sure about what actual threshold I would want to set here, my actual guess is that I wouldn’t want to end up funded >40% by capability labs, or their present employees.
- ^
Thank you so much! I really appreciate it.
It’s something I am still thinking about, but currently think that the framing of “FHI of the West” is probably the wrong framing. I didn’t yet know how to put what I am excited here into words, so didn’t put it into the plan section. I might write more about some of my thoughts here if I get around to it.
It’s a feature I built for last year’s fundraiser post!
If you send me source-code for a standalone Next.js component, or generic javascript widget, I can throw it up on the same server. I’ve done that one time for I think a METR post.
It would be cool to make it a generic offering, but allowing users to run arbitrary code on my server requires a lot of careful setup, and so would be a bunch of work.
Yep, just start a paragraph with >! and then you will be in a spoiler tag
Toss a bitcoin to your Lightcone – LW + Lighthaven’s 2026 fundraiser
No, the many cases we have of Sam lying to people. Including in this case the part where he seems to have lied to board members about what other board members believed.
Ah, that makes sense. I failed to parse it that way! Thanks for clarifying!
I suspect this is a growing problem with Lighthaven events.
Most of our events aren’t open access! I am also pretty confused who would show up to the afterparty to mock Solstice. What a weird move.
Yep, what they did seem pretty cowardly. I do also think it really wasn’t on the same level as what Sam did.
Promoted to curated: I do think this post summarizes one of the basic intuition generators for predicting AI motivations. It’s missing a lot of important stuff, especially as AIs get more competent (in-particular it doesn’t cover reflection, which I expect to be among the primary dynamics shaping motivations of powerful AIs), but it’s still quite helpful to be written up in more explicit form.
They were not, sorry about that!
Valuation of the top companies (based on a random Gemini summary when I searched for it, after a vague sense that things have gotten quite big).
“The prediction market”? I am confused what you mean by that.
Prediction markets broadly are now pretty economically important. In the last few years the industry has grown to ~$20B, which is not like, enormous, but pretty sizeable.
Huh, I never knew that term had any such narrowly construed meaning. Good to know! Given that, I agree that this is a bad name for the thing Ray is talking about.