I asked Forethought to apply. I think Forethought does great work, and am happy to support it.

Theory of change: I largely view Forethought's theory of change as doing high-level strategic thinking about what should be done in the world, given that we will plausibly see Transformative AI in the near future. What questions are people missing? How should we prioritize? What mistakes are we making? etc.

Empirically, their work on AI-enabled coups seems to have been somewhat influential, and important people at Anthropic and DeepMind think highly of them. So, I think they are well-placed to have an impact with this work.

They have a bunch of extremely smart and capable people, and I expect these people are well placed to think about fucky, confusing things and get somewhere.

The track record of this kind of work seems fantastic to me. I'm not confident there is still significant impact to be had, but it's plausible enough that it seems like it should clearly have a non-trivial investment.

I have not personally engaged much with Forethought's work (it's not super relevant to what I personally am doing, and I'm very behind on paper reading). So, this is less a personal endorsement based on an inside view, and more that this just seems obviously reasonable. Multiple people I respect seem to think well of Forethought.

Why me? Why am I, as a regranter, making this grant? I was a bit hesitant about this, but based on chatting to grantmakers it didn't seem like Forethought was likely to have its funding gap filled by large donors, yet both Open Philanthropy and Longview said positive things when I asked about giving to Forethought. For various reasons that I don't think are particularly important OpenPhil can't give money to Forethought directly, but still endorses them.

Funding gap: Forethought has a significant funding gap for their goal of 18 months runway at an aggressive growth rate, though they also have enough runway that I see this donation as marginally helping them be comfortable with an aggressive growth rate and increasing their runway, rather than something critical or deeply counterfactual. But I also think it is good to not insist on being the donor of last resort.

I also haven't found anything better to spend my regranting budget on, and so think I should err more towards funding things and away from procrastination / holding out for something better, especially with the prospect of a bunch of Anthropic employee funding in future. But that was a fairly marginal factor and I endorse the grant on the merits.

@Krishna-Patel I sent you a LinkedIn message but now I see your email is on your profile. Coming up

Thanks we're really excited about it too! What's the easiest way to coordinate a chat? @MarcusAbramovitch

This looks extremely promising to me. Want to have a 30 min call about it?

I'm very excited about this and it seems like a great value for 20k!

I donated $25k outside Manifund, for similar reasons as Daniel and Joel. I agree with Joel that the AI Village seems exciting from a scientific perspective in addition to a comms perspective (I don't feel very qualified to judge how good the comms side is, but I think the scientific upside is potentially large).

I'll add to what Joel wrote that AI Village can teach us things about model psychology / alignment in addition to their real-world capabilities. Seeing different models all work in the same environment on the same goal gives a lot of qualitative insight into their personality differences. One challenge that seems exciting to tackle is how to extract these insights into a useful form. My guess is that people who've spent a of time following the AI Village in detail have some interesting intuitive knowledge about model psychology already and it would be nice to disseminate this knowledge more widely somehow.

Recommended my remaining regrantor budget, $31.7k, for $71.7k total.

I continue to stand behind my last comment (and hopefully will write more later). I think that something like AI village is one of the most interesting science (not comms!!!) plays in AI today. I'm disappointed that this continues to be a non-consensus pick.

I believe that biological brains are VERY scalable. it is actually quite surprising that this cause is so underfunded

Forethought’s work has been some of the biggest influence on my safety research work, raising several new areas of research I and some of my collaborators hadn’t previously considered or prioritized. Keep up the amazing work!

see previous report, should have closed then

We have posted our 2025 retrospective and fundraiser post! I updated our project description with the text of it, and you can also read it on LessWrong: https://www.lesswrong.com/posts/eKGdCNdKjvTBG9i6y/toss-a-bitcoin-to-your-lightcone-lw-lighthaven-s-2026

If you're interested in joining us, go read our Request for Startups: https://seldonlab.com/rfs

@charbel-raphael

• Blocked a predatory investor that may have had a 40% chance to destroy an impactful mission 2 years down the line (the biggest risk)

• Counterfactually led to Workshop Lab's successful round, as reported independently by them

• Established and supported their re-incorporation as public benefit corporations, deploying the AIS mission in their legal registration

• Was counterfactually responsible for at least $565k direct investments into their rounds

• Direct design and product decisions supported by us, including pivots into more viable business models to support their mission

• Much more understanding for them of the investors' perspective and how to navigate them

• Hiring decisions and support through the Apart network, including running a 1,000-person hackathon

• New friends (pretty invaluable to have a community within which you can support each other with full trust - this keeps paying dividends now)

The counterfactual impact seems radically higher than an equivalent non-profit accelerator due to the scale of their downstream impact but the direct impact on them changing strategic directions was lower due to the founder demographic's strong (and most often good) opinions.

I started this because I've always hated politics, now, I am helping others openly learn civics & politics through bias-free game-based tools.

@niplav Still have to get my collaborators' data.

@saulmunn A quick n' dirty analysis automated via Claude on my six datapoints gives me 7.28 ± 2.89 hours of sleep after Orexin-A, 8.09 ± 0.75 after Placebo (i.e. Cohen's d of -0.386) at a p-value 0.556.

I've been following Dan's work for years and think this could go pretty far with agents.

The state of the art in collating followup studies is extremely poor for stupid reasons and so easy to beat.

He's not naive about the technical issues nor the likely reception in academia and has some cool ideas for getting the results into actual discussions.

@charbel-raphael

Hi Charbel,

I fundamentally disagree. Your assessment conflates passive reach with active comprehension.

You cannot build deep intuition for Reinforcement Learning or agentic behavior via TikTok, just as you cannot replace flight simulators with YouTube videos. There is a massive pedagogical difference between watching a creator talk about alignment and personally struggling against a misalignment failure mode in a simulation. The former creates views, the latter creates understanding.

Moreover, it is concerning that you are downvoting this project while citing your own competing venture as the "dominant" alternative. Given that your project roadmap implies seeking significant future funding to scale your specific strategy, aggressively downvoting alternative methodologies looks less like community moderation and more like an attempt to monopolize the "AI Awareness" vertical.

Solving the engagement gap requires a diverse portfolio of strategies (interactive games, simulations, and videos), not a zero sum fight to establish your personal medium as the only valid approach.

Congrats on getting the grant/ those numbers!

I wonder if it makes sense to translate those clips into a few languages automatically?

I personally think that this project deserved more. Those numbers are quite impressive, and the Transformer newsletter is a critical epistemic infrastructure for the community.

I think the numbers highlighted are pretty impressive (~$570 per GWWC pledge and ~$1700 per MARS fellow), and I think marketing is important and would benefit many organisations.

From what I see, GWWC estimates the average 10% pledger donates $100,000 over the course of their pledge, and they secured 2 pledges from the EA Netherlands campaign alone—so this is already worth it for this. I would give them money.

I asked Forethought to apply. I think Forethought does great work, and am happy to support it.

Theory of change: I largely view Forethought's theory of change as doing high-level strategic thinking about what should be done in the world, given that we will plausibly see Transformative AI in the near future. What questions are people missing? How should we prioritize? What mistakes are we making? etc.

Empirically, their work on AI-enabled coups seems to have been somewhat influential, and important people at Anthropic and DeepMind think highly of them. So, I think they are well-placed to have an impact with this work.

They have a bunch of extremely smart and capable people, and I expect these people are well placed to think about fucky, confusing things and get somewhere.

The track record of this kind of work seems fantastic to me. I'm not confident there is still significant impact to be had, but it's plausible enough that it seems like it should clearly have a non-trivial investment.

I have not personally engaged much with Forethought's work (it's not super relevant to what I personally am doing, and I'm very behind on paper reading). So, this is less a personal endorsement based on an inside view, and more that this just seems obviously reasonable. Multiple people I respect seem to think well of Forethought.

Why me? Why am I, as a regranter, making this grant? I was a bit hesitant about this, but based on chatting to grantmakers it didn't seem like Forethought was likely to have its funding gap filled by large donors, yet both Open Philanthropy and Longview said positive things when I asked about giving to Forethought. For various reasons that I don't think are particularly important OpenPhil can't give money to Forethought directly, but still endorses them.

Funding gap: Forethought has a significant funding gap for their goal of 18 months runway at an aggressive growth rate, though they also have enough runway that I see this donation as marginally helping them be comfortable with an aggressive growth rate and increasing their runway, rather than something critical or deeply counterfactual. But I also think it is good to not insist on being the donor of last resort.

I also haven't found anything better to spend my regranting budget on, and so think I should err more towards funding things and away from procrastination / holding out for something better, especially with the prospect of a bunch of Anthropic employee funding in future. But that was a fairly marginal factor and I endorse the grant on the merits.

@Krishna-Patel I sent you a LinkedIn message but now I see your email is on your profile. Coming up

Thanks we're really excited about it too! What's the easiest way to coordinate a chat? @MarcusAbramovitch

This looks extremely promising to me. Want to have a 30 min call about it?

I'm very excited about this and it seems like a great value for 20k!

I donated $25k outside Manifund, for similar reasons as Daniel and Joel. I agree with Joel that the AI Village seems exciting from a scientific perspective in addition to a comms perspective (I don't feel very qualified to judge how good the comms side is, but I think the scientific upside is potentially large).

I'll add to what Joel wrote that AI Village can teach us things about model psychology / alignment in addition to their real-world capabilities. Seeing different models all work in the same environment on the same goal gives a lot of qualitative insight into their personality differences. One challenge that seems exciting to tackle is how to extract these insights into a useful form. My guess is that people who've spent a of time following the AI Village in detail have some interesting intuitive knowledge about model psychology already and it would be nice to disseminate this knowledge more widely somehow.

Recommended my remaining regrantor budget, $31.7k, for $71.7k total.

I continue to stand behind my last comment (and hopefully will write more later). I think that something like AI village is one of the most interesting science (not comms!!!) plays in AI today. I'm disappointed that this continues to be a non-consensus pick.

I believe that biological brains are VERY scalable. it is actually quite surprising that this cause is so underfunded

Forethought’s work has been some of the biggest influence on my safety research work, raising several new areas of research I and some of my collaborators hadn’t previously considered or prioritized. Keep up the amazing work!

see previous report, should have closed then

We have posted our 2025 retrospective and fundraiser post! I updated our project description with the text of it, and you can also read it on LessWrong: https://www.lesswrong.com/posts/eKGdCNdKjvTBG9i6y/toss-a-bitcoin-to-your-lightcone-lw-lighthaven-s-2026

If you're interested in joining us, go read our Request for Startups: https://seldonlab.com/rfs

@charbel-raphael

• Blocked a predatory investor that may have had a 40% chance to destroy an impactful mission 2 years down the line (the biggest risk)

• Counterfactually led to Workshop Lab's successful round, as reported independently by them

• Established and supported their re-incorporation as public benefit corporations, deploying the AIS mission in their legal registration

• Was counterfactually responsible for at least $565k direct investments into their rounds

• Direct design and product decisions supported by us, including pivots into more viable business models to support their mission

• Much more understanding for them of the investors' perspective and how to navigate them

• Hiring decisions and support through the Apart network, including running a 1,000-person hackathon

• New friends (pretty invaluable to have a community within which you can support each other with full trust - this keeps paying dividends now)

The counterfactual impact seems radically higher than an equivalent non-profit accelerator due to the scale of their downstream impact but the direct impact on them changing strategic directions was lower due to the founder demographic's strong (and most often good) opinions.

I started this because I've always hated politics, now, I am helping others openly learn civics & politics through bias-free game-based tools.

@niplav Still have to get my collaborators' data.

@saulmunn A quick n' dirty analysis automated via Claude on my six datapoints gives me 7.28 ± 2.89 hours of sleep after Orexin-A, 8.09 ± 0.75 after Placebo (i.e. Cohen's d of -0.386) at a p-value 0.556.

I've been following Dan's work for years and think this could go pretty far with agents.

The state of the art in collating followup studies is extremely poor for stupid reasons and so easy to beat.

He's not naive about the technical issues nor the likely reception in academia and has some cool ideas for getting the results into actual discussions.

@charbel-raphael

Hi Charbel,

I fundamentally disagree. Your assessment conflates passive reach with active comprehension.

You cannot build deep intuition for Reinforcement Learning or agentic behavior via TikTok, just as you cannot replace flight simulators with YouTube videos. There is a massive pedagogical difference between watching a creator talk about alignment and personally struggling against a misalignment failure mode in a simulation. The former creates views, the latter creates understanding.

Moreover, it is concerning that you are downvoting this project while citing your own competing venture as the "dominant" alternative. Given that your project roadmap implies seeking significant future funding to scale your specific strategy, aggressively downvoting alternative methodologies looks less like community moderation and more like an attempt to monopolize the "AI Awareness" vertical.

Solving the engagement gap requires a diverse portfolio of strategies (interactive games, simulations, and videos), not a zero sum fight to establish your personal medium as the only valid approach.

Congrats on getting the grant/ those numbers!

I wonder if it makes sense to translate those clips into a few languages automatically?

I personally think that this project deserved more. Those numbers are quite impressive, and the Transformer newsletter is a critical epistemic infrastructure for the community.

I think the numbers highlighted are pretty impressive (~$570 per GWWC pledge and ~$1700 per MARS fellow), and I think marketing is important and would benefit many organisations.

From what I see, GWWC estimates the average 10% pledger donates $100,000 over the course of their pledge, and they secured 2 pledges from the EA Netherlands campaign alone—so this is already worth it for this. I would give them money.