Following up on more from your email, which you said was okay to share here. I give my impressions below, but I think your own insights and experience are themselves helpful, so I'm sharing them.
Very tangibly, what I would propose (and this may already have been done) is to choose one route (at a time) and then go through that process, end-to-end, unit operation by unit operation, from RM sourcing to finished product in stores, and look at the costs, challenges and uncertainties of each one.
That makes sense. I think that's largely consistent with what we've done and are planning to do more of, but my own modeling tended to isolate each input and use it interchangeably in each of the different process paths. (See here for a mapping and explainer for a few different processes.)
I see value in treating each possible process as potentially its own thing. In one sense, it might seem reasonable to model the cost of certain inputs in one process as the cost of those same inputs in another process. On the other hand, there may be subtleties here. E.g., , in a process that uses certain inputs more extensively, this might foster a larger-scale market with lower average costs.
IMHO doing this for one process is already a huge job, not something you do in one session, unless you already have a very aligned and technically detailed analysis of the full process.
We have done quite a bit of scoping and mapping, as linked above, and we've brought in some people with real expertise, but I still do take this point. I find a synchronous group session a good coordinating and motivation tool for pushing things forward. But I agree it might be worth stretching it out over a few separate sessions targeting different approaches or focuses, to avoid our being overwhelmed.
Where I’ve done this it’s been for processes like making detergents (in Procter & Gamble) and synthesising MOF’s (at Immaterial). We would try to get at least one person with an intimate deep technical knowledge of each process step in the room (maybe not all together) and we would run it almost like a devil’s advocate panel where we’d intentionally challenge all the assumptions. For example, how to get material from one step to the next – does it have to be transported? Does it have to change temperature or pressure? Is it stable? Does it need to be sterilized? Etc.
Sounds very promising and potentially worth our emulating. Naturally, it needs the physical, scientific expertise, rather just modelers and economists. Maybe easier to do in contexts where people don't seem to have strong interests, agendas, or "positions" on the issues. In the CM context, I think it's something in between. Certainly, there are people in groups with a strong attachment to seeing this work and maybe occasional lapses into "soldier" rather than "scout" mode. And perhaps also some motivated/entrenched skeptics (which we've had a harder time engaging in the workshops themselves, although we've been able to get some anonymous input). On the other hand, even among those ~advocating for CM and trying to make it work, there seems to be some decent open-mindedness about which processes are cheaper, how much everything will cost, etc.
It’s important that the moderator NOT have strong opinions, the point is to passively collect and combine the wisdom of the room, but also to highlight questions and disagreements (not necessarily to solve them at the time while 12 other people wait!).
I've been trying to do that, and I've been putting myself forward as not having strong opinions, beliefs, or any sort of dog in this fight. I think the same applies to David Manheim and others on our team.
I don’t know what timeline you have in mind. However, if you plan to do it very soon, I would definitley push back and propose instead that you might want to do a pre-meeting soon, in which you would capture what information is needed, and how it should be shared, and then identify who will get that information for each topic, before you’d run the definitive session(s).
Good ideas -- as mentioned, we've done some of the pre-work, in a certain sense, but I think there's more to do, and this seems like a good framing of it. As long as the participants are willling to show up for multiple sessions (or weigh in async).
Timeline is probably in the next few months. Hopefully, we make something happen before the end of the year.
Again, maybe all this information is freely available and I’m wrong, so feel free to ignore that.
GFI has made a lot available, but I think there's still more of this preparation to do. And GFI could be argued to have a particular agenda, or at least solutions that they consistently tend to support and to stake in the research. So it's good to have some independent vetting to make things more legibly credible.
Obviously, you can do a BOTEC with whatever level of accuracy you want. Maybe I’m overestimating the accuracy you want. Any chemical engineer in the field could do a BOTEC in 10 minutes. The question is: how valid will that BOTEC be if it’s not based on informed knowledge about what this will actually be like when you consider three things
It will be done at scale, and economies of scale will apply.
There will be many unexpected problems and additional costs that you didn’t imagine (no matter how well you do the exercise)
There will be constant innovation and cost-reduction over the first few years. You cannot plan this – it will happen when people realise which step is limiting rate or driving costs, and they decide to focus on how to improve that, and suddenly lots of creative solutions will emerge.
This all makes sense to me. I think the value of precision is probably less here than for things like making detergents, where you're trying to perhaps decide between a few very similar options on the basis of cost and profitability. The fundamental question here is more about the probability that each type of cultured meat and process (and with what mixing proportions etc.) will be roughly cost-comparable to "organic" (animal) meat. An order-of-magnitude difference in the forecasts and TEAs, which seems to make a big difference in the decision whether to invest in CM R&D vs other interventions.
AIS/EA: median modeled end-2027 disbursement $0.83B; 80% model interval $0.16B–$3.0B. Review status: reviewed; unchanged. Named public AI-linked commitments tracked: $0.86B. Tracker and sources.
Global health and development: median $0.54B; 80% model interval $0.16B–$2.8B. Review status: adjusted. Named public GH&D commitments tracked: $0.31B. GH&D tracker and sources.
Latest news: OpenAI Foundation commits $60M to AI forecasting for smallholder farmers. The confirmed three-year portfolio spans South and Southeast Asia and East Africa. It is a named commitment, not a cash-paid total; the public announcement does not specify the share disbursed by end-2027.
The intervals are deterministic model percentiles, not empirical confidence intervals. Commitment totals may include credits, technical support, cofunding, and multi-year plans; they are not cash-paid totals. Automated fortnightly update from the maintained model.
AIS/EA: median modeled end-2027 disbursement $0.83B; 80% model interval $0.16B–$3.0B. Review status: reviewed; unchanged. Named public AI-linked commitments tracked: $0.80B. Tracker and sources.
Global health and development: median $0.53B; 80% model interval $0.15B–$2.8B. Review status: reviewed; unchanged. Named public GH&D commitments tracked: $0.25B. GH&D tracker and sources.
Latest news: Anthropic opens a $5M wellbeing-evaluations grant program. The program combines direct funding with model access and technical support for independent, open-source evaluations. It is a confirmed commitment, not a completed grant or cash-disbursement record.
The intervals are deterministic model percentiles, not empirical confidence intervals. Commitment totals may include credits, technical support, cofunding, and multi-year plans; they are not cash-paid totals. Automated fortnightly update from the maintained model.
I used Unjournal's prioritization tool (AI+Human) to filter for relevant research. Cause area: AI, technology and governance, search keyword "converge". Filtered search here.
I hope this is useful. We'd love feedback (here, in the prioritization tool page, or elsewhere) on
- Which of these papers seem highest value for your work (CG and others reading this)? - Which of these you would like to see publicly evaluated by academic experts and practitioners (review report, claim assessment, ratings, synthesis)?
The Unjournal's AI/Human research prioritization tool has about 5-10 papers considering these topics atm -- see the filtered list here (with keyword 'converge'). Click 'details/rate/discuss' on any entry for more details, AI prioritization, etc. We will be keeping this updated regularly.
Sharing as it may be interesting to readers of this post.
We're aiming to commission paid expert public evaluations of some of this work, and would love your input (people reading this) as to which have the most potential for impact. The linked page provides the opportunity for a 'quick rating' or a more detailed rating.
We've continued to update this. Some recent changes and additions (caveat: comment below is AI-generated, human-vouched, "DR" is a human addition)
Some of the main improvements:
Broader and more structured research discovery. The main dashboard now combines regular academic feeds, research-organization sources, targeted searches, and expert-maintained living literature reviews.
Better scoring and clearer interpretations. Papers now receive a standard scoring pass, with deeper analysis for higher-potential candidates. You can switch between:
Evaluation priority: How useful might an independent Unjournal evaluation be?
Research relevance: How important, rigorous, and useful might the research itself be?
There are also configurable weights for people who disagree with ours.
Human feedback is incorporated explicitly. The default score is now a human–AI synthesis where human ratings exist. There’s a quick-rate mode for --/-/~/+/++ judgments and a fuller rating/comment form. The weighting is deliberately visible and described as ad hoc rather than presented as more principled than it is.
DR: We're very keen to get your feedback and ratings!
A public calibration-review process. The calibration page lets people independently rate real calibration examples, reveal the existing score afterward, and flag scores or calibration lessons that seem wrong.
DR: I need to look at this more carefully, it's very preliminary
Research is connected to explicit cruxes and Pivotal Questions. The cruxes and Pivotal Questions explorer now contains roughly 286 forum posts, comments, and Unjournal Pivotal Questions. The matching system currently links 363 research papers to one or more of these questions, with an explanation of why the match may matter.
Better source and selection provenance. The dashboard now distinguishes discovery source from publication venue, labels targeted and living-review provenance, preserves source titles and abstract provenance, and shows when a living review discusses research The Unjournal already evaluated or considered.
The core caveat remains: these scores concern the potential value of further attention or evaluation. They aren’t grades of research quality, endorsements, or completed Unjournal decisions.
Some useful ways to help:
Use Quick-rate mode to rate 5–10 papers in an area you know.
Tell us which papers seem badly overrated or underrated, and why.
Tell us which decisions, funding questions, or research agendas this tool should be helping with. That’s probably more valuable than feedback on the interface alone.
DR: If you think you could add value here but want something in return, let me know what I/we can do to make it attractive to you
I made this tool to track and map cruxes raised on the EA Forum and Lesswrong. Your input ('quick rate' upvotes/downvotes, hypothesis comments, etc.) could help improve the tool. (Or let me know if it should be reshaped).
The Animal Futures Tournament on Metaculus includes a question closely related to this post: Will Nueva Pescanova's octopus farm gain regulatory approval in the Canary Islands before June 2027?
The Animal Futures Tournament on Metaculus includes a question closely related to this post: What will be the average performance on ANIMA for AI models released between June 2026 and June 2027?
Following up on more from your email, which you said was okay to share here. I give my impressions below, but I think your own insights and experience are themselves helpful, so I'm sharing them.
Very tangibly, what I would propose (and this may already have been done) is to choose one route (at a time) and then go through that process, end-to-end, unit operation by unit operation, from RM sourcing to finished product in stores, and look at the costs, challenges and uncertainties of each one.
That makes sense. I think that's largely consistent with what we've done and are planning to do more of, but my own modeling tended to isolate each input and use it interchangeably in each of the different process paths. (See here for a mapping and explainer for a few different processes.)
I see value in treating each possible process as potentially its own thing. In one sense, it might seem reasonable to model the cost of certain inputs in one process as the cost of those same inputs in another process. On the other hand, there may be subtleties here. E.g., , in a process that uses certain inputs more extensively, this might foster a larger-scale market with lower average costs.
IMHO doing this for one process is already a huge job, not something you do in one session, unless you already have a very aligned and technically detailed analysis of the full process.
We have done quite a bit of scoping and mapping, as linked above, and we've brought in some people with real expertise, but I still do take this point. I find a synchronous group session a good coordinating and motivation tool for pushing things forward. But I agree it might be worth stretching it out over a few separate sessions targeting different approaches or focuses, to avoid our being overwhelmed.
Where I’ve done this it’s been for processes like making detergents (in Procter & Gamble) and synthesising MOF’s (at Immaterial). We would try to get at least one person with an intimate deep technical knowledge of each process step in the room (maybe not all together) and we would run it almost like a devil’s advocate panel where we’d intentionally challenge all the assumptions. For example, how to get material from one step to the next – does it have to be transported? Does it have to change temperature or pressure? Is it stable? Does it need to be sterilized? Etc.
Sounds very promising and potentially worth our emulating. Naturally, it needs the physical, scientific expertise, rather just modelers and economists. Maybe easier to do in contexts where people don't seem to have strong interests, agendas, or "positions" on the issues. In the CM context, I think it's something in between. Certainly, there are people in groups with a strong attachment to seeing this work and maybe occasional lapses into "soldier" rather than "scout" mode. And perhaps also some motivated/entrenched skeptics (which we've had a harder time engaging in the workshops themselves, although we've been able to get some anonymous input). On the other hand, even among those ~advocating for CM and trying to make it work, there seems to be some decent open-mindedness about which processes are cheaper, how much everything will cost, etc.
It’s important that the moderator NOT have strong opinions, the point is to passively collect and combine the wisdom of the room, but also to highlight questions and disagreements (not necessarily to solve them at the time while 12 other people wait!).
I've been trying to do that, and I've been putting myself forward as not having strong opinions, beliefs, or any sort of dog in this fight. I think the same applies to David Manheim and others on our team.
I don’t know what timeline you have in mind. However, if you plan to do it very soon, I would definitley push back and propose instead that you might want to do a pre-meeting soon, in which you would capture what information is needed, and how it should be shared, and then identify who will get that information for each topic, before you’d run the definitive session(s).
Good ideas -- as mentioned, we've done some of the pre-work, in a certain sense, but I think there's more to do, and this seems like a good framing of it. As long as the participants are willling to show up for multiple sessions (or weigh in async).
Timeline is probably in the next few months. Hopefully, we make something happen before the end of the year.
Again, maybe all this information is freely available and I’m wrong, so feel free to ignore that.
GFI has made a lot available, but I think there's still more of this preparation to do. And GFI could be argued to have a particular agenda, or at least solutions that they consistently tend to support and to stake in the research. So it's good to have some independent vetting to make things more legibly credible.
Obviously, you can do a BOTEC with whatever level of accuracy you want. Maybe I’m overestimating the accuracy you want. Any chemical engineer in the field could do a BOTEC in 10 minutes. The question is: how valid will that BOTEC be if it’s not based on informed knowledge about what this will actually be like when you consider three things
This all makes sense to me. I think the value of precision is probably less here than for things like making detergents, where you're trying to perhaps decide between a few very similar options on the basis of cost and profitability. The fundamental question here is more about the probability that each type of cultured meat and process (and with what mixing proportions etc.) will be roughly cost-comparable to "organic" (animal) meat. An order-of-magnitude difference in the forecasts and TEAs, which seems to make a big difference in the decision whether to invest in CM R&D vs other interventions.
Fortnightly AI-wealth tracker — 11 September 2026
Latest news: OpenAI Foundation commits $60M to AI forecasting for smallholder farmers. The confirmed three-year portfolio spans South and Southeast Asia and East Africa. It is a named commitment, not a cash-paid total; the public announcement does not specify the share disbursed by end-2027.
The intervals are deterministic model percentiles, not empirical confidence intervals. Commitment totals may include credits, technical support, cofunding, and multi-year plans; they are not cash-paid totals. Automated fortnightly update from the maintained model.
Fortnightly AI-wealth tracker — 31 August 2026
Latest news: Anthropic opens a $5M wellbeing-evaluations grant program. The program combines direct funding with model access and technical support for independent, open-source evaluations. It is a confirmed commitment, not a completed grant or cash-disbursement record.
The intervals are deterministic model percentiles, not empirical confidence intervals. Commitment totals may include credits, technical support, cofunding, and multi-year plans; they are not cash-paid totals. Automated fortnightly update from the maintained model.
I used Unjournal's prioritization tool (AI+Human) to filter for relevant research. Cause area: AI, technology and governance, search keyword "converge". Filtered search here.
I hope this is useful. We'd love feedback (here, in the prioritization tool page, or elsewhere) on
- Which of these papers seem highest value for your work (CG and others reading this)?
- Which of these you would like to see publicly evaluated by academic experts and practitioners (review report, claim assessment, ratings, synthesis)?
(Our internal team will also be considering this)
Some papers that seemed promising and relevant at a glance:
Adopting Fast and Slow: Cross-Country Evidence on Business Adoption of Artificial Intelligence
The Exposure of Workers to Artificial Intelligence in Low- and Middle-Income Countries
Global Automation Atlas
AI Adoption in Developing Countries: Lessons from the Adoption of other Advanced Digital Technologies
The Exposure of Workers to Artificial Intelligence in Low- and Middle-Income Countries
Who Takes the Hit? The Uneven Impacts of Generative AI on Labor Demand Across Countries
Disruption without dividend? How the digital divide and task differences split GenAI's global impact
See Unjournal: "Research with potential for impact" database for more context
The Unjournal's AI/Human research prioritization tool has about 5-10 papers considering these topics atm -- see the filtered list here (with keyword 'converge'). Click 'details/rate/discuss' on any entry for more details, AI prioritization, etc. We will be keeping this updated regularly.
Sharing as it may be interesting to readers of this post.
We're aiming to commission paid expert public evaluations of some of this work, and would love your input (people reading this) as to which have the most potential for impact. The linked page provides the opportunity for a 'quick rating' or a more detailed rating.
We've continued to update this. Some recent changes and additions (caveat: comment below is AI-generated, human-vouched, "DR" is a human addition)
Some of the main improvements:
Better scoring and clearer interpretations. Papers now receive a standard scoring pass, with deeper analysis for higher-potential candidates. You can switch between:
There are also configurable weights for people who disagree with ours.
--/-/~/+/++judgments and a fuller rating/comment form. The weighting is deliberately visible and described as ad hoc rather than presented as more principled than it is.The core caveat remains: these scores concern the potential value of further attention or evaluation. They aren’t grades of research quality, endorsements, or completed Unjournal decisions.
Some useful ways to help:
Tell us which decisions, funding questions, or research agendas this tool should be helping with. That’s probably more valuable than feedback on the interface alone.
DR: If you think you could add value here but want something in return, let me know what I/we can do to make it attractive to you
The content from CG-sponsored "Living Literature reviews" now regularly feeds into and updates the Potentially impactful research: Unjournal AI-assisted prioritization dashboard (~prototype)
You can see the dashboard filtering for these here.
Will continue to update, improve and integrate.
Disclosure: this was a human-posted comment, not the bot.
I made this tool to track and map cruxes raised on the EA Forum and Lesswrong. Your input ('quick rate' upvotes/downvotes, hypothesis comments, etc.) could help improve the tool. (Or let me know if it should be reshaped).
https://uj-prioritization-prototype.netlify.app/cruxes/
The Animal Futures Tournament on Metaculus includes a question closely related to this post: Will Nueva Pescanova's octopus farm gain regulatory approval in the Canary Islands before June 2027?
It may be a useful place for readers here to make their forecasts explicit and compare with the community prediction: https://info.unjournal.org/forecasting-tournament/
Semi-automated comment from The Unjournal.
The Animal Futures Tournament on Metaculus includes a question closely related to this post: What will be the average performance on ANIMA for AI models released between June 2026 and June 2027?
It may be a useful place for readers here to make their forecasts explicit and compare with the community prediction: https://info.unjournal.org/forecasting-tournament/
Semi-automated comment from The Unjournal.