EA Forum Podcast (All audio)

Scaling inference

With the release of OpenAI's o1 and o3 models, it seems likely that we are now contending with a new scaling paradigm: spending more compute on model inference at run-time reliably improves model performance. As shown below, o1's AIME accuracy increases at a constant rate with the logarithm of test-time compute (OpenAI, 2024).

OpenAI's o3 model continues this trend with record-breaking performance, scoring:

• 2727 on Codeforces, which makes it the 175th best competitive programmer on Earth;
• 25% on FrontierMath, where "each problem demands hours of work from expert mathematicians";
• 88% on GPQA, where 70% represents PhD-level science knowledge;
• 88% on ARC-AGI, where the average Mechanical Turk human worker scores 75% on hard visual reasoning problems.

According to OpenAI, the bulk of model performance improvement in the o-series of models comes from increasing the length of chain-of-thought (and possibly further techniques like "tree-of-thought") and improving the chain-of-thought [...]

---

Outline:

(02:52) AI safety implications

(04:04) AGI timelines

(04:55) Deployment overhang

(06:12) Chain-of-thought oversight

(07:45) AI security

(09:10) Interpretability

(10:07) More RL?

(11:31) Export controls

(11:57) Conclusion

The original text contained 4 footnotes which were omitted from this narration.

---

First published:

January 15th, 2025

Source:

https://forum.effectivealtruism.org/posts/bYXjejHrvq65jFL9s/implications-of-the-inference-scaling-paradigm-for-ai-safety

---

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.

Are you ready to make a real difference in the world through research? The AIM Research Program (ARP) empowers you to help address some of the world's most pressing challenges with rigorous, evidence-based research. This fully funded, 12-week program provides the skills, mentorship, and tools you need to produce impactful, decision-relevant work.

Applications are now open for the next edition of the AIM Research Program, running from April 7 to June 27, 2025. If you’re passionate about impact-focused research, this is your chance to join our program. Applications close on February 2nd, 2025.

Apply now

What is the AIM Research Program?

The AIM Research Program is a 12-week, fully funded initiative designed to train exceptional researchers in impact-driven methodologies. Participants will:

• Work on high-impact research projects with expert mentorship.
• Contribute to decision-influencing reports on global health, animal welfare, and other priority areas.
• Develop critical skills in analysis, problem-solving, and [...]

---

Outline:

(01:02) What is the AIM Research Program?

(01:56) Program Details

(02:19) Who Should Apply?

(02:52) What We Offer

(03:34) What about AIM's other programs?

(04:09) How to Apply

---

First published:

January 15th, 2025

Source:

https://forum.effectivealtruism.org/posts/SAgDds3PDndh59heQ/announcing-applications-for-the-aim-research-program-april

---

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.

I am no prophet, and here's no great matter.

— T.S. Eliot, “The Love Song of J. Alfred Prufrock”

This post is a personal account of a California legislative campaign I worked on March-June 2024, in my capacity as the indoor air quality program lead at 1Day Sooner. It's very long—I included as many details as possible to illustrate a playbook of everything we tried, what the surprises and challenges were, and how someone might spend their time during a policy advocacy project.

History of SB 1308 Advocacy Effort

SB 1308 was introduced in the California Senate by Senator Lena Gonzalez, the Senate (Floor) Majority Leader, and was sponsored by Regional Asthma Management and Prevention (RAMP). The bill was based on a report written by researchers at UC Davis and commissioned by the California Air Resources Board (CARB). The bill sought to ban [...]

---

Outline:

(00:40) History of SB 1308 Advocacy Effort

(18:22) Key Takeaways

(21:40) Personal Reflections

The original text contained 2 footnotes which were omitted from this narration.

---

First published:

January 15th, 2025

Source:

https://forum.effectivealtruism.org/posts/pTSqg6MNjNsMPsTT8/retrospective-on-the-california-sb-1308-campaign

---

Narrated by TYPE III AUDIO.

Now accepting applications for Code4Compassion 2025!

Join leading AI developers & animal advocates to build practical tech solutions for animal protection in this 24-hour event developed in collaboration between AI for Animals, Electric Sheep, & Open Paws.

• 4 problem tracks based on real-world technical needs submitted by animal advocacy orgs
• Utilise the first-ever suite of AI models specifically trained for animal advocacy: Pre-trained & fine-tuned models, along with dedicated cloud compute for each team & expert mentorship
• Pro-animal, pro-innovation: Strategically positioned between EAGx Bay Area (Feb 21-23) & the AI for Animals 2025 Conference (March 1-2), C4C25 creates a direct pipeline from technical innovation to real-world deployment

Whether you’re an AI engineer, full-stack developer, ML researcher, or data scientist, if you’re excited about applying your skills to protect animals, we want to hear from you.

• Date: February 26-27, 2025
• Location: Sports Basement Hacking Space, 1590 Bryant Street [...]

---

First published:

January 14th, 2025

Source:

https://forum.effectivealtruism.org/posts/Duu5JQLuJqntK4gyi/announcing-code4compassion-2025-a-hackathon-transforming

---

Narrated by TYPE III AUDIO.

Until recently

I

thought Julia and I were digging a bit into savings to

donate more. With the tighter funding climate for effective altruism

we

thought

it was worth spending down a bit, especially considering that our

expenses should decrease significantly in 1.5y when our youngest

starts kindergarten.

I was surprised, then, when I ran the

numbers and realized that despite donating 50% of a reduced

income, we were $9k (0.5%) [1] richer than when I left

Google two years earlier.

This is a good problem to have! After thinking it over for the last

month, however, I've decided to start earning

less: I've asked for a voluntary salary reduction of $15k/y

(10%). [2] This is something I've been thinking about off and on since

I started working at a non-profit: it's much more efficient

to reduce your salary than it is to make a donation. [...]

---

First published:

January 15th, 2025

Source:

https://forum.effectivealtruism.org/posts/3TLTrJS2DZJ5mcrkc/voluntary-salary-reduction

---

Narrated by TYPE III AUDIO.

The views expressed here are my own, not those of people who provided feedback.

Summary

• I estimate the cost-effectiveness of paying farmers to use more humane pesticides to decrease the suffering of wild insects is 236 DALY/$, or:
  • 23.7 times the cost-effectiveness of GiveWell's top charities.
  • 141 times the cost-effectiveness of broiler welfare campaigns.
  • 51.4 times the cost-effectiveness of cage-free campaigns.
  • 36.9 % of the past cost-effectiveness of the Shrimp Welfare Project (SWP).
• I guess research on and advocacy for more humane pesticides are way more cost-effective than paying farmers to use them more. For example, I estimate cage-free campaigns improve 408 times as many chicken-years per $ as buying organic instead of barn eggs. So I guess research on and advocacy for more humane pesticides is more cost-effective than SWP has been.
• I can easily see my estimated cost-effectiveness going up or down by a few orders of magnitude. I guess uncertainty [...]

---

Outline:

(00:14) Summary

(02:16) Calculations

(06:30) Discussion

(09:15) Acknowledgements

The original text contained 6 footnotes which were omitted from this narration.

---

First published:

January 14th, 2025

Source:

https://forum.effectivealtruism.org/posts/mgsiDB2Kkm3mDSWWP/cost-effectiveness-of-paying-farmers-to-use-more-humane

---

Narrated by TYPE III AUDIO.

Alt Title: A Case for Supporting the Yes On IP28 Campaign

It is safe to say, that from the middle of January, 1877, until the following October, the most prominent theme of public discussion was this question of suffrage for women. Miners discussed it around their camp-fires, and “freighters” on their long slow journeys over the mountain trails argued pro and con, whether they should “let” women have the ballot.

History of Woman Suffrage, Volume 3

Introduction

Denigrated as “bawling, ranting women, bristling for their rights,” the all-male electorate voted two-to-one rejecting the 1877 ballot measure that would have extended suffrage to the women of Colorado. This was the fourth women's suffrage ballot initiative to be defeated at the polls, after previous campaigns in Michigan (1874), Nebraska (1871), and Kansas (1867) received similarly unfavorable results—securing only 23%, 22%, and 31% of the vote, respectively. Nevertheless, they persisted. Between [...]

---

Outline:

(00:38) Introduction

(05:54) Commence the Conversation

(14:50) Build a Bloc

(20:56) Advocate with Authenticity

(26:49) Conclusion

---

First published:

January 10th, 2025

Source:

https://forum.effectivealtruism.org/posts/sNQ7HdcjYEfvM5WXA/how-can-we-get-the-world-to-talk-about-animals-inviolable-1

---

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 post is my personal perspective. I’m sure that my colleagues on the Forum Team and at CEA disagree with parts of this. However, since I am the Interim EA Forum Project Lead, I recognize that my opinions and beliefs carry extra weight. I’m very happy to receive feedback and push back from others, since I believe that my decisions matter a fair amount. You’re welcome to reply to this post, DM me, find me at EAG Bay Area, contact our team, or leave our team anonymous feedback here.

When I took the role of Interim EA Forum Project Lead in late August 2024, I spent some time investigating where the Forum was at and thinking about what (if anything) our team should prioritize working on. Over the course of 2024 (and indeed, since early 2023), Forum usage metrics have steadily gone down[1]. My subjective opinion was that the [...]

---

Outline:

(01:21) The Forum Team as community builders

(05:41) What does the best version of the Forum community  look like?

(07:23) We're not there yet

(09:50) What is the Forum Team doing?

(12:01) What are we not doing?

(13:00) How you can help

(14:31) Appendix: The value of the Forum

The original text contained 27 footnotes which were omitted from this narration.

---

First published:

January 13th, 2025

Source:

https://forum.effectivealtruism.org/posts/wpDGEXjAtHJa2eCFA/what-are-we-doing-about-the-ea-forum-jan-2025

---

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.

Poker pro, art collector, photographer, investor, AI researcher, chronic website creator, endless traveller, and omnipresent volunteer in nascent things. An independent and an invariant.

I briefly worked with him on an accountability partner service. We had funding but he never invoiced me. Every time I called him he was somewhere else on Earth. Senegal, Israel, Nepal, Egypt. He spent 13 straight months travelling in 2017-8.

He wasn't much of a writer - you won't find him on here - but he had started. What suddenly turned out to be his final projects were Poker Camp, Hold'LLM, and Bet Mitzvah, an unwritten book on probability and instrumental reason.

Here are some pieces about him from people who knew him much better than me. I expect there to be more.

• https://andrew.gr/stories/chisness/
• https://x.com/chisness
• https://blog.rossry.net/chisness/
• https://redeniusfuneralhomes.com/obituary/max-chiswick/
• https://forumserver.twoplustwo.com/29/news-views-gossip/remembering-life-max-chiswick-aka-chisness-legacy-far-beyond-poker-tables-1844405/
• https://oldjewishmen.substack.com/p/bhif-old-jewish-men-loses-a-friend 

His last commit was on the 22nd December. He died of malaria on [...]

---

First published:

January 13th, 2025

Source:

https://forum.effectivealtruism.org/posts/r9fJ26ca5cneY3hA8/max-chiswick-1985-2025

---

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 Otherwise. Most EAs won't find these arguments new.

Last month, Emma Goldberg wrote a NYT piece contrasting effective altruism with approaches that refuse to quantify meaningful experiences. The piece indicates that effective altruism is creepily numbers-focused. Goldberg asks “what if charity shouldn’t be optimized?”

The egalitarian answer

Dylan Matthews gives a try at answering a question in the piece: “How can anyone put a numerical value on a holy space” like Notre Dame cathedral? For the $760 million spent restoring the cathedral, he estimates you could prevent 47,500 deaths from malaria.

“47,500 people is about five times the population of the town I grew up in. . . . It's useful to imagine walking down Main Street, stopping at each table at the diner Lou's, shaking hands with as many people as you can, and telling them, ‘I think you need to die to make a cathedral [...]

---

Outline:

(00:29) The egalitarian answer

(01:16) Who prefers magnificence?

(03:10) Inequality has its benefits

(04:34) Is there enough for everybody to have access to the finer things?

(05:37) The balance of good and bad

(06:33) Both sides have ugly aspects

(07:04) These aren't the only choices

(08:58) Related:

The original text contained 1 footnote which was omitted from this narration.

---

First published:

January 13th, 2025

Source:

https://forum.effectivealtruism.org/posts/TiFeCBxKj79bohoDY/the-ugly-sides-of-two-approaches-to-charity

---

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.