EA Forum Podcast (All audio)

The text of the bill can be found here. It begins by citing the warnings of AI company CEOs and deep learning pioneers Geoffrey Hinton and Yoshua Bengio, as well as the 2023 FLI open letter calling for a 6-month pause. The bill would prohibits the construction or upgrading of AI datacenters until Congress pass an AI safety law aimed at preventing AI companies "from releasing harmful products into the world that threaten the health and well-being of working families, our privacy and civil rights, and the future of humanity". It would also impose export controls for advanced chips to any country "to any country or entity that does not have laws and regulations in place to protect humanity from AI safety concerns and existential risks, protect workers, and protect the environment". Sen. Sanders and Rep. Ocasio-Cortez announced the bill in a live press conference:

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

The Community Building Grants (CBG) program, run by the Groups team at CEA since 2018, is undergoing substantial changes.

Through the CBG program, we fund a set of professional city and national EA groups[1] and support their leaders through retreats, regular check-ins and calls, a Slack space, and resources for running groups.

CBG groups have achieved exciting things, such as helping EAs navigate (policy) hubs, supporting hundreds of community members in taking next steps, and incubating new organisations.

In this post, we want to share what's changing and the reasoning behind it. In 2026, two things are changing:

• CBG grant evaluation is moving from the Groups team to EA Funds (which became part of CEA in the summer of 2025), where it will be managed alongside, but remain distinct from, the EA Infrastructure Fund. 
• CBG non-monetary support has been transitioned, integrated with other offers, or phased out.

Why we are making these changes

In 2025, we reviewed how the CBG program fits into CEA's broader strategy. As CEA aims to reach and raise EA's ceiling, we are shifting internal focus more towards scalable products that can support this ambition. After a long strategic process, we concluded that [...]

---

Outline:

(01:15) Why we are making these changes

(03:01) Whats changing

(05:11) What has not changed

(06:23) Trade-offs

(07:42) Finally

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

This post was inspired by debate week, but I also published it to my Substack. There, most of my readers are not as familiar with EA discourse, so I added a long introduction which I haven’t copied here (but which you might find entertaining). There's also a voiceover of the full version available on Substack or in any podcast feed if you search “Sandcastles”.

1. The heart of the matter

There's a debate taking place on the EA forum this week. The motion is:

If AGI goes well for humans, it’ll probably (>70% likelihood) go well for animals.

Taken literally, this could leave a 30% chance of AI going catastrophically bad for animals even if it went well for humans– think factory farms orbiting space colonies.

Many people working frantically on AI alignment agree with the statement “If AGI takes off hard and fast, it will probably (>70% likelihood) go well for humans.” 30% is still a terrifyingly high chance of disaster, more than enough to throw out everything else and focus on averting it.

Multiplying these two .7s gets us to .49, i.e. a less than 50% chance of AI going well for animals. But I don’t think [...]

---

Outline:

(00:35) 1. The heart of the matter

(03:25) 2. The make AI go well movement

(05:53) 3. AW and GHD are part of MAIGW

(07:29) 4. A new hope

(08:57) 5. How to be a TAI-focused AW movement

(11:39) 6. Raising an animal-friendly god 🚨 \*CRUX ALERT\*

(20:03) 7. How to be an animal-friendly Make AI Go Well movement

(24:34) 8. Family reunion

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Cross-posted from my website.

Epistemic status: Speculating with no central thesis. This post is less of an argument and more of a meditation.

A decade ago, before there was a visible path to AGI and before AI alignment was a significant research field, I figured the solution to the alignment problem would look something like Coherent Extrapolated Volition. I figured we'd find a way to get the AI to internalize human values. I had problems with this approach (why only human values?), but I still felt reasonably confident that the coherent extrapolation of human values would include concern for the welfare of all sentient beings. The CEV-aligned AI would recognize that factory farming is wrong, and that wild animal suffering is a big problem.

Today, the dominant research paradigms in AI alignment have nothing to do with CEV, and I don't know what to think.

Regarding the promisingness of today's popular research paradigms, my beliefs are aligned (heh) with those of most MIRI researchers: namely, I don't think they have promise. For example, see On how various plans miss the hard bits of the alignment challenge by Nate Soares. I'm not an alignment researcher, but [...]

The original text contained 8 footnotes which were omitted from this narration.

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Summary

In many worlds, but not all

- If AGI is widely available, alt protein advocates could direct it themselves and prioritization is not an issue.

- If AGI capability is concentrated in governments and companies, cultivated meat competes with other issues like cancer, climate, and national security.

- An autonomous AGI reasoning from first principles might prioritize it, but animal welfare is underrepresented in alignment frameworks—there is not yet a strong basis for assuming it would.

Very Likely

- Current AI tools are already highly capable of solving biological problems. AGI would likely be able to solve remaining scientific questions in cultivated meat.

- The science of scaling would also likely be handled well by AGI

- The data scarcity problem in alt proteins may be less of an issue with powerful, generalisable AGI.

Likely

- AGI could optimize facility design, supply chains, and input costs, improving economics and attracting private capital.

- Physically building out an industry that barely exists takes time regardless of how good the designs are, but this may not meaningfully slow deployment.

Lean no

- AGI can [...]

---

Outline:

(00:13) Summary

(01:06) Intro

(02:18) For AGI to solve cultivated meat, all of the following must hold

(02:24) Cultivated meat (or animal welfare) is prioritized

(05:18) The remaining science is solved

(07:15) Manufacturing scales affordably

(09:29) Regulators speed up, maybe by adopting AI

(13:07) Political opposition reverses

(15:21) Consumers accept cultivated meat

(18:43) All of the above happen in coordination, on a timeline that matters for animals

(20:54) Caveats:

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

Epistemic status: medium, unsure if this is biased to what I have experienced, but think it is useful regardless.

I think that organisations and individuals should try be more transparent about their capacity, e.g. at conferences when people ask for volunteering / collaboration opportunities, a lot of orgs/individuals say yes, and then nothing ever really happens.

For example, some members from EA Bath attended EAGx Amsterdam last year. One of my members was pretty excited about EA and using his skills to help in climate change / GHD. He met with a lot of people who offered him opportunities and collaboration, so he followed up with them all after the conference, but it resulted in them all ghosting him, which lead to him being less sure that EAG/ EAGx's were a good use of his time. I've had similar experiences too, with asking people if they have opportunities and them saying yes, and then I receive no reply when I email them.

I think that these people / organisations are genuinely excited for more people like me (students / entry-level) to get involved, but are also probably time constrained. Because of this I think it's super important that [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

TL;DR

I spent eight months as a community builder for an EA group. This is a reflection on what that work actually looks from the inside, why I’ve found it's structurally undersupported, and what I think could be different for funders, organisations, and for community builders themselves.

What community builders actually do

From the outside, community building can look deceptively simple. Organising events, running fellowships, hosting discussions. And yes, those things are part of it. But most of the actual work happens in the spaces between. The one-on-one conversation with someone trying to figure out how they want to contribute. The introduction you make between two people who probably should've met a year ago. The moment someone finally feels comfortable asking the question they've been sitting with for months because the room feels safe enough.

These things don't show up cleanly in metrics. But they tend to be the moments that actually shape people's trajectories.

If you think about how someone typically moves toward an impact-focused career or project, it rarely happens in one step. There's an initial spark, be it through an event, an article, or even a conversation. Then, there's deeper engagement. After that, usually what comes [...]

---

Outline:

(00:13) TL;DR

(00:33) What community builders actually do

(02:03) The call that made me write this

(03:17) The strange place community building holds

(04:41) What doesnt show up in a grant application

(07:12) Whats actually broken, and what could be different

(11:08) Communities arent background noise

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

TL;DR

AI safety organizations are capacity constrained by a lack of senior researchers who can mentor and supervise junior talent. This bottleneck drives hyper-selective hiring, blocks promising candidates from entering the field, and is being exacerbated by AI automation, which is raising the bar for human contribution rather than lowering it. Based on 23 interviews with hiring managers, research leads, and funders across the ecosystem, we identify six concrete actions field-builders can take to improve talent pipelines, including standardized mentor evaluations, direct policy embedding, and production codebase access.

Study Context

In Q4, 2025, MATS Research performed 23 interviews with hiring managers, research leads, and funders across the AI safety ecosystem to understand talent needs, hiring constraints, and skills gaps. This analysis focuses on identifying how field-building organizations like MATS Research can better prepare candidates for employment in this rapidly evolving field. We are sharing this information more broadly in hopes it helps organizations design programs that better prepare people to contribute to AI safety work.

Methodology and Limitations

Orgs were selected according to MATS Research's understanding of the AI Safety ecosystem, drawing from a range of categories (funders, think tanks, labs, nonprofits and for-profit safety organizations as well as [...]

---

Outline:

(00:13) TL;DR

(00:54) Study Context

(01:31) Methodology and Limitations

(02:20) A Rapidly Growing Field Induces Hyper-Selectivity

(04:12) Two Distinct Tracks at Frontier AI Companies

(05:20) The Automation Paradox

(06:27) Evolving Demand Patterns

(07:19) Critical Skills Gaps

(09:21) Actionable Recommendations

(11:50) The Path Forward

(12:21) Acknowledgements

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

This report was mostly written by Claude Opus 4.6. We manually checked all claims and didn’t find any errors.

Summary

• We summarise the results of a survey carried out before the 2026 Summit on Existential Security. Respondents were Summit attendees: leaders and key thinkers in the x-risk and AI safety communities.
• The survey asked attendees about their estimates of existential risk, AGI timelines, and resource allocation priorities.

• Survey data comes from the 59 respondents who consented to their answers being shared publicly.[1] Data was collected in February 2026.

Median: 25%

Mean: 34%

Median: 2033

Mean: 2034

Median: 2030

Mean: 2030

We defined AGI as:

An AI system (or collection of systems) that can fully automate the vast majority (>90%) of roles in the 2025 economy. A job is fully automatable when machines could be built to carry out the job better and more cheaply than human workers. Think feasibility, not adoption.

Resource allocation

• The strongest consensus in resource allocation is [...]

---

Outline:

(00:23) Summary

(01:00) X-risk and timelines

(01:39) Resource allocation

(02:25) Areas of debate and consensus at the event

(02:51) Existential risk estimates

(03:07) Summary statistics

(03:25) Distribution

(04:27) Notable comments

(05:52) AGI timelines

(06:35) Summary statistics

(06:59) Distribution of AGI estimates

(07:54) Notable comments

(08:46) Relationship between timelines and risk

(09:25) Resource allocation priorities

(09:51) Overall preferences

(10:29) Misaligned AI views by timeline subgroup

(11:05) Notable comments

(12:04) Sub-field priorities

(12:55) Notable comments

(13:59) Areas of consensus and debate at the event

(14:08) Areas of broad consensus

(15:09) Areas of active debate

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

OpenAI is now a public benefit corporation, with a charter that demands they use AGI for the benefit of all, and do so safely. To justify this structure to the Attorneys General of Delaware and California, they split off the nonprofit OpenAI Foundation, and instead of full ownership, gave it 27% equity, worth well over $150 billion - what some have called the largest theft in human history. They also convened a commission to advise them on how to give away that money, and last year announced the first tranche of that giving, evidently funded with their equity.

This week, they announced a team, and an even larger commitment; giving "at least $1 billion" over the coming year; I argue everyone should agree that this is far too little.

The plan is to use their massive endowment, currently over $150 billion, for charity - though technically, it's not an endowment. Which is convenient, because if it was, they would need to give 5% of their assets every year, currently over $7 billion per year. But even that seems quite conservative, given the possible trajectory of their holdings.

It instead seems like there are three relevant possibilities:

• [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.