EA Forum Podcast (All audio)

We’ve recently published a set of design sketches for AI tools that help with collective epistemics.

We think that these tools could be a pretty big deal:

• If it gets easier to track what's trustworthy and what isn’t, we might end up in an equilibrium which rewards honesty
• This could make the world saner in a bunch of ways, and in particular could give us a better shot at handling the transition to more advanced AI systems

We’re excited for people to get started on building tech that gets us closer to that world. We’re hoping that our design sketches will make this area more concrete, and inspire people to get started.

The (overly-)specific technologies we sketch out are:

• Community notes for everything — Anywhere on the internet, content that may be misleading comes served with context that a large proportion of readers find helpful
• Rhetoric highlighting — Sentences which are persuasive-but-misleading, or which misrepresent cited work, are automatically flagged to readers or writers
• Reliability tracking — Users can effortlessly discover the track record of statements on a given topic from a given actor; those with bad records come with health warnings
• Epistemic [...]

---

First published:

Source:

Linkpost URL:
https://www.forethought.org/research/design-sketches-collective-epistemics

---

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.

“I” refers to Zach, the Centre for Effective Altruism's CEO. Oscar is CEA's Chief of Staff. We are grateful to all the CEA staff and community members who have contributed to the development and implementation of our strategy. Mistakes are of course our own.

One year into our 2025-26 strategy, we have reversed our programs’ negative growth trajectory from 2023-24 and will sustain that momentum in 2026 while preparing to hit much more ambitious goals from 2027 onwards:

• CEA grew the number of people engaging with our programs by 20–25% year-over-year across each tier of our engagement funnel, beating our targets of 7.5–10% without increasing spending, and reversing the moderate decreases in engagement with our programs throughout 2023–24.
• We laid the foundations for furthering our contribution to EA funding diversification by merging with EA Funds and hiring an accomplished new Director (Loic Watine).
• And we strengthened CEA's own foundations, establishing our in-house Operations Team and growing our headcount from 42 to 66 while increasing talent density, including another incoming experienced Director for our new Strategy and M&E function (Rory Fenton).

We also faced some challenges:

• Beyond the impact of EA growth on the perception of the [...]

---

Outline:

(00:34) Executive summary

(03:05) Stewardship and sustainable momentum

(04:22) Growing the EA community

(05:05) 2025

(08:17) 2026

(09:35) Improving the EA brand

(10:02) 2025

(15:24) 2026

(16:30) Diversifying EA funding

(16:50) 2025

(19:53) 2026

(21:20) Strengthening CEA

(21:38) Operations Team

(22:07) Staffing

(23:38) Spending

(24:46) Monitoring & evaluation

(25:40) EV, still

---

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: I built a simple back-of-the-envelope model of AI agent economics that combines Ord's half-life analysis of agent reliability with real inference costs. The core idea is that agent cost per successful outcome scales exponentially with task length, while human cost scales linearly. This creates a sharp viability boundary that cost reductions alone cannot meaningfully shift. The only parameter that matters much is the agent's half-life (reliability horizon), which is precisely the thing that requires the continual learning breakthrough (which I think is essential for AGI-level agents) that some place 5-20 years away. I think this has underappreciated implications for the $2T+ AI infrastructure investment thesis.

The setup

Toby Ord's "Half-Life" analysis (2025) demonstrated that AI agent success rates on tasks decay exponentially with task length, following a pattern analogous to radioactive decay. If an agent completes a 1-hour task with 50% probability, it completes a 2-hour task with roughly 25% probability and a 4-hour task with about 6%. There is a constant per-step failure probability, and because longer tasks chain more steps, success decays exponentially.

METR's 2025 data showed the 50% time horizon for the best agents was roughly 2.5-5 hours (model-dependent) and had been doubling every ~7 [...]

---

Outline:

(00:57) The setup

(02:04) The model

(03:26) Results: base case

(05:01) Finding 1: cost reductions cannot beat the exponential

(06:24) Finding 2: the half-life is the whole game

(08:02) Finding 3: task decomposition helps but has limits

(09:33) What this means for the investment thesis

(11:38) Interactive model

(11:57) Caveats and limitations

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

TLDR: People plot benchmark scores over time and then do math on them, looking for speed-ups & inflection points, interpreting slopes, or extending apparent trends. But that math doesn’t actually tell you anything real unless the scores have natural units. Most don’t.

Epistemic status: I haven’t vetted this post carefully, and have no real background in benchmarking or statistics.

Benchmarks look like rulers; they give us scores that we want to treat as (noisy) measurements of AI progress. But since most benchmark score are expressed in quite squishy units, that can be quite misleading.

• The typical benchmark is a grab-bag of tasks along with an aggregate scoring rule like “fraction completed”[2]

• ✅ Scores like this can help us...

  • Loosely rank models (“is A>B on coding ability?”)

  • Operationalize & track milestones (“can [...]

---

Outline:

(01:00) Benchmark scores vs units of AI progress

(02:42) Exceptions: benchmarks with more natural units

(04:48) Does aggregation help?

(06:27) Where does this leave us?

(06:30) Non-benchmark methods often seem better

(07:32) Mind the Y-axis problem

(09:05) Bonus notes / informal appendices

(09:13) I. A more detailed example of the Y-axis problem in action

(11:53) II. An abstract sketch of whats going on (benchmarks as warped projections)

---

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.

Considering how important politics are to most EA cause areas, I find it surprising how little attention political work gets among EAs. I get the feeling that because politics are blocked from the Forum homepage, and because EA fellowships fail to mention high-leverage political work, young EAs often assume that this is not a high impact space without looking into it themselves.

Luckily, this is changing.

There is a growing movement of EAs recognizing the huge importance of politics, and realizing it's not as intractable as many of us assume. Over the past few months, we have joined together with EA club leaders from schools like Middlebury, the University of Michigan, Princeton, and MIT to create multiple curriculum/content options for EA groups to use exploring this cause area. What we've put together:

• One-week fellowship curriculum - easily added on to your existing intro or advanced fellowships! Can also be used to guide a reading-based general meeting.
• Three-week fellowship curriculum - for a deeper dive into the importance, tractability, and neglectedness of democracy preservation & U.S. politics.
• Slides presentation - for a presentator-led meeting on democracy preservation, with discussion & activities built in. 

Link to content. [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

The USDA New World Screwworm Grand Challenge represents the largest single funding opportunity for screwworm research and control technologies in decades. Applications are due February 23, 2026.

What's happening?
The New World Screwworm (NWS) is a flesh-eating fly endemic to the Americas that lays eggs in open wounds of warm-blooded animals. The larvae burrow into living flesh, consuming the host from inside. Infected animals experience excruciating pain, stop eating, and die slowly over days to weeks. This parasite occasionally infects humans too. NWS was previously eradicated from North and Central America, although containment efforts failed in 2024 and the parasite has reached Northern Mexico in January 2026.

New technologies like gene-drive-enhanced sterile insect technique (SIT) could make large-scale control and eradication faster and cheaper than ever before. The USDA Grand Challenge represents an opportunity to direct funding to developing technologies with potential applications beyond the United States southern border. Next-gen SIT and gene-drive could help extend the range of control and eradication to South America for the first time, safeguarding rural livelihoods and the wellbeing of hundreds of millions of animals.

Screwworm Free Future is supporting qualified applicants -- particularly those working in or with endemic regions [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

I've been reading Toby Ord's recent sequence on AI scaling a bit. General notes come first, then my thoughts.

Notes

---

Outline:

(00:20) Notes

(00:26) Takeaways

---

First published:

Source:

Linkpost URL:
https://www.dissent.blog/notes-on-toby-ords-ai-scaling-series/

---

Narrated by TYPE III AUDIO.

Bruce Friedrich's new book, MEAT, is a deeply thoughtful, pragmatic, and hope-inspiring story about alternative proteins. In his role as the Founder and President of the Good Food Institute, Bruce is uniquely positioned to give an insider's account of how far plant-based, fermentation-derived, and cultivated meat have come, and where they are likely to go from here.

The book's publication comes at what seems to be a critical junction for alternative proteins. Several years of exponential growth in sales of plant-based meat and investment in cultivated meat have given way to a period of uncertainty for both. MEAT is a timely and compelling reminder that the fundamental reasons why these innovations are needed are as relevant today as they have ever been. More than that, it is a case for optimism.

In the first section, Friedrich makes an extremely thorough and well-evidenced case for meat alternatives in terms of global food security, climate change, antibiotic resistance, and pandemic prevention. As a vegan who is totally on board with the moral case for alternative proteins, I found myself waiting for the chapter containing grizzly descriptions of factory farming, but it never came.

The book is the opposite of sanctimonious. [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Over 100 billion farmed fish are slaughtered each year. At Coefficient Giving, where I (Michelle Lavery) work as a Senior Program Associate, we estimate that only ~0.5% of them are reliably stunned before slaughter.

For the more than a trillion wild-caught fish killed annually, conditions are even worse: most are left to suffocate slowly in air or in low-oxygen water, a process that can take minutes to hours.

Though fish slaughter represents only a few hours in an animal's life, we believe the pain it causes is excruciating — especially when done poorly, which is most of the time. This problem is both urgent and tractable: the suffering is immense, the moment of slaughter is discrete and identifiable, and people are genuinely horrified when they learn about current practices. So why hasn't it been solved? The answer lies in a combination of technical complexity, limited competition in the equipment market, and a critical skills gap between the biologists who study the problem and the engineers who could solve it.

That's why we're launching this RFP: to bridge that gap and catalyze the engineering innovation this problem demands.

Why this is hard

If you've never thought about how fish [...]

---

Outline:

(01:24) Why this is hard

(03:43) Who/what were looking for

(04:57) How to apply

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Evolutionary cost-balancing arguments in welfare biology (the study of animal well-being in the wild) assume that producing suffering and happiness has metabolic or neural costs, and that natural selection allocated these states efficiently. A specific version of this reasoning – which I'll call the "Evening Out Argument" – holds that if a type of suffering is highly likely and largely unavoidable for an animal, evolution would reduce its intensity. The logic is that, if suffering functions to motivate learning and avoidance, then intense suffering provides little behavioral benefit when the animal can barely avoid the bad outcome anyway – it's just cost without payoff.

I argue this logic fails for large categories of animal experience. Most notably, it doesn't apply to background motivational states like hunger or anxiety, which function differently from discrete learning signals. I'll also discuss the prevalence of chronic and maladaptive suffering, concerns about the biological plausibility of some assumptions behind the Evening Out Argument, and why high infant mortality doesn't imply reduced suffering in long-lived species.

For context, many people familiar with the topic have explicitly stated that they do not consider cost-balancing arguments to be strong; however, I have seen some people [...]

---

Outline:

(01:30) A brief history

(06:12) Mayflies vs baby turtles

(08:41) Hedonic accounting: discrete events vs background states

(13:14) Chronic suffering and old age

(15:27) Other biological plausibility concerns: path dependencies and modularity

(17:36) Ancestral environments and environmental mismatch

(19:11) Concluding thoughts and implications

(21:44) Appendix: A puzzle for Evening Out

(24:26) Acknowledgments

The original text contained 32 footnotes which were omitted from this narration.

---

First published:

Source:

---

Narrated by TYPE III AUDIO.