EA Forum Podcast (All audio)

This is a link post.Photo by gibblesmash asdf

Self-driving cars are not a solved problem, nor are they close to getting solved. I will explain further, but since this is a case where the messenger matters more than the message, first listen to Andrej Karpathy tell you the same thing.

Karpathy is as credentialed as it comes: he was the lead AI researcher responsible for the development of Tesla's Full Self-Driving software from 2017 to 2022. (Karpathy also did two stints as a researcher at OpenAI, taught a deep learning course at Stanford, and coined the term “vibe coding”.)

Here's a long quote from Karpathy's October 17, 2025 interview with Dwarkesh Patel:

Dwarkesh Patel 01:42:55

You’ve talked about how you were at Tesla leading self-driving from 2017 to 2022. And you firsthand saw this progress from cool demos to now thousands of cars out there actually autonomously doing drives. Why did that take a decade? What was happening through that time?

Andrej Karpathy 01:43:11

One thing I will almost instantly push back on is that this is not even near done, in a bunch of ways that I’m going to get to. Self-driving is very interesting because it's definitely where [...]

---

Outline:

(08:08) Do self-driving car companies believe self-driving cars are a solved problem?

(20:18) The major caveats to Waymo's success

(26:29) Broader implications

---

First published:

December 28th, 2025

Source:

https://forum.effectivealtruism.org/posts/9zLByNJkAvqdShNwq/self-driving-cars-aren-t-nearly-a-solved-problem

Linkpost URL:
https://strangecosmos.substack.com/p/self-driving-cars-arent-nearly-a

---

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.

TLDR: I'm leaving San Francisco to do policy in DC within a few months. As the lead organizer of EA SF, it's likely the group will go dormant if no one takes on this role. However, EA SF is uniquely situated to be a potentially very high impact city group, due to its location in an AI development hub with a high concentration of EA-adjacent people. Thus, becoming lead organizer would be extremely counterfactually impactful - no one else would do it if not you. Furthermore, this role doesn't require much work (a minimal version can be done with less than 5h/wk). If you're located in San Francisco, Berkeley, or the Peninsula, and you're interested in helping out, please reach out.

EA SF's Theory of Change

Attention from within the EA community itself has shifted away from EA qua EA and towards pure AI safety. So why run an EA group in 2026?

There are a few reasons (which I partially borrow from https://forum.effectivealtruism.org/posts/R8AAG4QBZi5puvogR/effective-altruism-in-the-age-of-agi):

1. You can incentivize work on neglected topics. There are several that would be ignored by pure AI safety groups trying to be legible and respectable to the broader public, and don't really have a [...]

---

Outline:

(00:58) EA SFs Theory of Change

(04:32) Challenges

(05:40) What The Role Entails

(05:56) Minimal

(06:34) Maximal

(07:21) Interested?

---

First published:

January 5th, 2026

Source:

https://forum.effectivealtruism.org/posts/mQoq3RXtwhgNwLBGN/ea-san-francisco-needs-a-new-lead-organizer

---

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.

Stanford EA is calling for aid!

This fall we worked to revive Stanford EA but now our co-president is taking a leave of absence for a research fellowship and we could use help keeping momentum.

Presently, we've got:

• 2 organizers,
• ~12 interested members
• Great support from CEA
• Materials, social accounts, co-working rooms, etc. 

Our biggest bottleneck is finding organizers, especially a co-president. If you are at Stanford and interested in EA we'd love to chat with you. We think there is a clear vision and path to impact, if potentially interested but unsure please reach out and let us pitch you!

Other ways to help:

• If you know anyone at Stanford interested in EA or that we should otherwise reach out to
• Stanford EA has many great alumni, we'd love to be able to point to concrete outcomes and impact for past members. If you are a previous Stanford alumni okay with being featured on our alumni spotlight (Website buildout in progress) that would be a big help!

You can reach us by DM or email: [email protected]

---

First published:

January 5th, 2026

Source:

https://forum.effectivealtruism.org/posts/DdoHPu6smhwoBJQ4t/stanford-ea-is-calling-for-aid

---

Narrated by TYPE III AUDIO.

This is a link post.

Summary: This post argues that Artificial General Intelligence (AGI) threatens both liberal democracy and rule-based international order through a parallel mechanism. Domestically, if AGI makes human labor economically unnecessary, it removes the structural incentive for inclusive democratic institutions—workers lose leverage when their contribution is no longer essential. Internationally, if AGI gives one nation overwhelming productivity advantages, it erodes other countries' comparative advantages, reducing the benefits of trade and weakening incentives to maintain a rule-based world order. The post draws historical parallels to early 20th century concerns about capital concentration, distinguishes between "maritime" (trade-dependent) and "continental" (autarkic) power strategies, and discusses what middle powers like the EU might do to remain relevant. The core insight is that both democracy and international cooperation rest on mutual economic dependence—and AGI could eliminate both dependencies simultaneously.

Read this if you're interested in: AGI's geopolitical implications, how economic structures shape political systems, the future of liberal democracy, or strategic options for countries that won't lead in AGI development.

Epistemic status: fairly speculative and likely incomplete or inaccurate, though with a lot of interesting links.

Introduction

The Effective Altruism community has long acknowledged the risks of AGI, especially those related to [...]

---

Outline:

(01:49) Introduction

(04:03) Leverage without labour

(07:08) The maritime world order

(10:20) Pax silica

(12:52) A normative aside on transactional alliances

(14:38) A Roadmap for the AI Middle Powers

(19:41) Conclusion

---

First published:

January 1st, 2026

Source:

https://forum.effectivealtruism.org/posts/NsGREdfrJbdvGKwLp/agi-and-the-structural-foundations-of-democracy-and-the-rule

Linkpost URL:
https://substack.com/home/post/p-183131518

---

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 spent the last few days building a tool called Aligned (thealignedapp.com), a web app designed to map where opinions converge and diverge. It uses binary polling (Yes/No/Not Sure) to track how opinions change over time, measure agreement rates between users, and compare human and AI alignment. I am sharing it here to get feedback on whether a tool like this is actually useful for the community.

Motivations

I built this project to explore three specific problems regarding how we form and track opinions:

1. A "Metaculus" for Values: The EA community has excellent tools like Metaculus and Manifold for tracking predictions. However, I feel we lack an equivalent product for tracking opinions and values. I wanted to build something that could serve that function—a place to log where we stand on issues that aren't necessarily verifiable predictions but are still crucial for coordination.

2. Clarity Regarding Beliefs: Disagreements often feel total, but they are usually specific. I wanted a way to sort questions by "Most Split" to identify exactly where a community is divided rather than relying on intuition. The app allows for "Anonymous Voting" and "Anonymous Posting", which I hope encourages users to reveal their true [...]

---

Outline:

(00:13) Summary

(00:40) Motivations

(02:23) How it Works & Development Context

(03:39) Relevance to Effective Altruism

(04:16) Limitations and Uncertainties

(04:54) Feedback Requested

---

First published:

January 3rd, 2026

Source:

https://forum.effectivealtruism.org/posts/Rb4HmkapRtDtFWNyj/seeking-feedback-a-tool-for-opinion-value-tracking-and

---

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 spent the last few days building a tool called Aligned (thealignedapp.com), a web app designed to map where opinions converge and diverge. It uses binary polling (Yes/No/Not Sure) to track how opinions change over time, measure agreement rates between users, and compare human and AI alignment. I am sharing it here to get feedback on whether a tool like this is actually useful for the community.

Motivations

I built this project to explore three specific problems regarding how we form and track opinions:

1. A "Metaculus" for Values: The EA community has excellent tools like Metaculus and Manifold for tracking predictions. However, I feel we lack an equivalent product for tracking opinions and values. I wanted to build something that could serve that function—a place to log where we stand on issues that aren't necessarily verifiable predictions but are still crucial for coordination.

2. Clarity Regarding Beliefs: Disagreements often feel total, but they are usually specific. I wanted a way to sort questions by "Most Split" to identify exactly where a community is divided rather than relying on intuition. The app allows for "Anonymous Voting" and "Anonymous Posting", which I hope encourages users to reveal their true [...]

---

Outline:

(00:13) Summary

(00:40) Motivations

(02:23) How it Works & Development Context

(03:15) Relevance to Effective Altruism

(03:52) Limitations and Uncertainties

(04:30) Feedback Requested

---

First published:

January 3rd, 2026

Source:

https://forum.effectivealtruism.org/posts/Rb4HmkapRtDtFWNyj/seeking-feedback-a-tool-for-opinion-value-tracking-and

---

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.

Note: Until recently, I worked on comms and content at Giving What We Can. The opinions expressed here are my own.

If I learned one thing during my ~2.5 years at Giving What We Can, it's that talking about effective giving (in a way that resonates with new audiences) is harder than it seems.

In light of that, I wanted to share an example of an ad that popped up in my feed recently because I think the messaging is particularly strong.

Above: Photo collage of the carousel ad. (Viewer swipes to see each of the six screens.)

Why I think this is strong messaging

1. It speaks to one of the personas effective giving orgs need to target: the “confused altruist”.  When I think about who might be actively searching for effective giving, a specific type of person comes to mind — I've named them the "confused altruist". This is a person who is naturally caring and feels some responsibility to help, but also isn’t really sure how. They are likely overwhelmed by the world's problems and distressed by the lack of concrete avenues for change. This person will notice the sheer number of donation opportunities [...]

---

Outline:

(00:57) Why I think this is strong messaging

(06:10) Why Im sharing these thoughts

---

First published:

January 2nd, 2026

Source:

https://forum.effectivealtruism.org/posts/BbndfpAaEbupH7kSz/great-bit-of-messaging-from-the-life-you-can-save

---

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.

This is a link post.

Exaggerating the risks (Part 20: AI 2027 timelines forecast, benchmarks and gaps)

When some of the issues with the time horizons forecast were pointed out, the AI 2027 authors have defended themselves by pointing out they actually did two models, and the time horizon model that we have discussed so far is a simplified one that they do not prefer. When you use their preferred model, the “benchmark + gaps” model, the assumptions of the time horizon model are not as important. I disagree with this defence. In fact, I think that method 2 is in many ways a worse model than method 1 is. I think in general, a more complicated model has to justify [its] complications, and if it doesn’t you end up in severe danger of accidentally overfitting your results or smuggling in the answer you want. I do not believe that model 2 justifies its complications.

Titotal, "A deep critique of AI 2027's bad timeline models"

Listen to this post

1. Introduction

This is Part 20 of my series Exaggerating the risks. In this series, I look at some places where leading estimates of existential risk look to have been exaggerated.

[...]

---

Outline:

(00:29) Exaggerating the risks (Part 20: AI 2027 timelines forecast, benchmarks and gaps)

(01:36) 1. Introduction

(02:33) 2. Model outline

(03:34) 3. RE-Bench saturation

(05:22) 4. Milestones and gaps

(06:11) 5. Speedups and simulation

(07:46) 6. Models and forecasts

(12:44) 7. The case of the missing model

(15:22) 8. Justifying the forecasts: A case study

(18:13) 8.3. Feedback loops: Eli's forecast

(29:12) 9. Speedups aren't the main problem

(30:46) 10. Conclusion

---

First published:

January 1st, 2026

Source:

https://forum.effectivealtruism.org/posts/ctoszPMhs9ZYBeZBk/exaggerating-the-risks-part-20-ai-2027-timelines-forecast

Linkpost URL:
https://reflectivealtruism.com/2025/08/08/exaggerating-the-risks-part-20-ai-2027-timelines-forecast-benchmarks-and-gaps/

---

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've heard that the end of the year is a good time to talk about the end of the world. AGI doom means different things to different people, so I thought it would be useful to get some idea what EAs think about it and what their probability of doom given AGI (P(doom)) is.

In a survey of thousands of AI experts, they think there is a wide variety of goodness/badness of AGI outcomes. However, some people think the outcome is very binary, extinction or bliss. In figure 10, I call these the MIRI flashlights (torches) because they are dark bars on the bottom with only bright above. That survey was not set up for longtermists, many of whom regard a scenario of disempowerment by AGI with rapid increases in human welfare as an existential catastrophe as it would be a drastic and permanent curtailment of the potential of humanity. Also, it did not disaggregate disempowerment and extinction.

Please select the mildest scenario that you would still consider doom. If you make a comment, it would be appreciated if you indicate your choice so others don’t have to search in the keys. If there is gradual disempowerment and [...]

---

First published:

December 29th, 2025

Source:

https://forum.effectivealtruism.org/posts/cScGFyq5WEEMFN3D8/quick-polls-on-agi-doom

---

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.

We invest heavily in fellowships, but do we know exactly where people go and the impact the fellowships have? To begin answering this question I manually analyzed over 600 alumni profiles from 9 major late-stage fellowships (fellowships that I believe could lead directly into a job following). These profiles represent current participants and alumni from MATS, GovAI, ERA, Pivotal, Talos Network, Tarbell, Apart Labs, IAPS, and PIBBS.

Executive Summary

• I’ve compiled a dataset of over 600 alumni profiles of 9 major 'late stage' AI Safety and Governance Fellowships.
• I found over 10% of fellows did another fellowship after their fellowship. This doesn’t feel enormously efficient.
• Almost ⅓ of ERA and Talos Network fellows (29.8% and 32.3% respectively) did another fellowship before or after, much higher than the average of 21.5%.
• ERA particularly seemed to be a ‘feeder’ fellowship for other fellowships. Only 9.5% of ERA fellows had done a fellowship before ERA, but 20.2% did another fellowship following, almost double the 11.1% average.
• GovAI Fellowship had strong direct links with other governance fellowships - i.e. many people went directly to or from other governance fellowships to GovAI. There were 13, 9 and 7 direct links between GovAI and ERA, IAPS and Talos Network respectively.
• [...]

---

Outline:

(00:43) Executive Summary

(02:08) Key Insights from mini-project

(02:57) Individual fellowships

(04:28) Links between fellowships

(05:40) Preliminary Directional Signals: IRG Data

(06:49) Notes on the Data

(08:18) Open Questions: What can this dataset answer?

(09:45) The Dataset Project

(10:21) Next Steps

---

First published:

January 2nd, 2026

Source:

https://forum.effectivealtruism.org/posts/SJBBgupFx7SXyBj2B/where-do-ai-safety-fellows-go-analyzing-a-dataset-of-600

---

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.