Tl;dr

• I saw Kabir's post on LessWrong and thought the argument could be strengthened by a framework from the pro-democracy field, which sorts defections into breaking (leave, visibly and publicly) and binding (stay and work from the inside). They can be further divided into the actions of speaking, acting, and standing in the way.
• Research by Dr. Jonathan Pinckney at the University of Texas at Dallas and Claire Trilling at SNF Agora Johns Hopkins on "defections" from authoritarian regimes or during democratic backsliding shows that "noncooperation" is ~2x as effective as simply "speaking out."
• "Loyalty shifts" are rare (23 out of 140 chances in their data). When they happen, they come from "quiet outreach" (~39% of cases), not protest (which did worse than no campaign at all).
• In this post, I'll attempt to translate this research to something meaningful for frontier lab employees, with the obvious caveat that a frontier lab is not a country or regime.
• Refusing the specific work while staying is the highest leverage action in their data. Quietly moving colleagues, and holding your position so someone who'll say yes to the employer doesn't get it, is likely the highest value action you could take.

Background

I've spent [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Applications to AIM's first animal welfare-focused incubation round close this Sunday, 13 September. If you are on the fence, this is my final nudge.

I almost didn’t apply

I only did because someone from the Ambitious Impact team nudged me six years ago. So at the risk of being annoying, I'll say it once more: the best way to find out if this is the right path for you is to apply. The selection process is built to help both of us work out whether you’re a match.

You don’t have to be an animal person

You don’t have to identify as an animal person, an activist, or a vegan. You can be, and plenty of excellent founders are, but it isn’t the bar. I’ve been in and out of veganism since 2017, and I run this program. The bar is taking seriously that factory farming is one of the biggest problems of our time, and being willing to follow the evidence on how to fix it. Animal people are just as welcome, and the best cohort will have a real mix of backgrounds.

Why now

We think the animal welfare movement is at a genuinely exciting moment. [...]

---

Outline:

(00:25) I almost didn't apply

(00:46) You don't have to be an animal person

(01:18) Why now

(01:48) Nobody is the perfect fit

(02:29) What you get

(02:51) What I'm hoping comes out of this round

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

As a daily reader of the EA Forum, I have found the experience significantly degraded recently. When I come to the /allPosts page, I am confronted with an onslaught of posts with a single upvote, mostly by brand-new users.

I want to give genuine new users a chance and read their stuff, and maybe upvote it. But when I see a pattern of "brand new user wrote long complicated post that currently has no upvotes", most of the time it's fully AI-generated, and it's a waste of my time to read it.

Even just reading through the titles wears me out and makes me want to leave. There will be maybe 15–20 posts for the day, most of which have long, kind of vague titles that take me a minute to parse.

In principle, I don't object to using AI to clean up your prose. But as far as I can tell, these posts usually have zero human thought behind them.*

Many of these posts are already automatically flagged as AI by Pangram. Rather than flagging and leaving them, these posts should be deleted. Or at minimum, the "AI-generated" flag should be visible from the [...]

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Prepared by: Claudinéia Pereira Costa, PhD; Meghan Barrett, PhD

Note: This post is very similar to our own prior EA Forum report on the genetic basis for nociception in another hemipteran, the lac bug (Kerria lacca, Hemiptera: Kerridae). In several places, we directly use the same text as we wrote in the first report to save time and ensure all limitations/caveats of the project are clearly, consistently reported. Consider this an attribution statement to our prior work for any directly copied text found within this post. All results for D. coccus are new and the methodology was slightly altered due to the genomic/proteomic resources available for this species.

Note: This report has not been peer-reviewed and has only been generated for the Forum. Results should be interpreted with caution, though we do follow peer-reviewed methodologies.

Executive Summary

• We analyzed the publicly-available cochineal bug Dactylopius coccus (Hemiptera: Dactylopiidae) genome, which was generated from sessile adult females, for evidence of twelve receptor genes in ten orthologous gene families related to heat, cold, chemical, and mechanical nociception in other insects.
  • Finding any of these genes does not prove nociception is present; functional assays and confirmation of these receptors in nociceptors would be [...]

---

Outline:

(01:14) Executive Summary

(03:13) Introduction

(07:44) Findings and Discussion

(07:48) Strong support for genes that may encode responsiveness to noxious heat, chemical, and mechanical stimuli in adult female D. coccus genome

(09:10) Noxious cold-sensing channel Pkd2, with function known from fruit flies, was not detected; but, other mechanisms for cold-sensing could still be present, as cold-sensing is critically understudied in arthropods

(11:03) Methods

(14:59) Acknowledgements

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

I wrote this post for High Impact Professionals and wanted to share it here since HIP now have an entrepreneurship track in their program, and applications for the AIM program are due September 13! Thanks a lot to @Danielle MacLean for encouraging me to write this.

What it's like to launch a nonprofit

A quick note before we start. This piece focuses on one part of the decision to start or join a nonprofit: what founding looks like day to day and who is a good personal fit for it. In addition to this, whether your idea is worth pursuing and what impact it could have also matter a lot, and I've written about those separately.

When deciding if you should start a nonprofit vs. join one, most people frame this as a question about ideas, skills or impact: is my idea good enough to build an organisation around, or should I lend my skills to one that already exists? That framing misses what matters just as much. Founding is a different job from any role inside an established org, and the deciding factor is rarely the idea. It is whether the shape of a founder's life fits the [...]

---

Outline:

(00:27) What it's like to launch a nonprofit

(02:00) You can choose your own hours (as long as it's all of them)

(03:54) A nonprofit founder is the ultimate generalist

(05:06) With founding, there is no manager and no map

(06:20) What founding costs: money, benefits, and recognition

(08:04) The nonprofit founder role changes as the organisation grows

(09:27) You cannot always walk away from founding as easily

(10:13) Career capital of founding vs. joining

(11:56) So who is founding for?

(14:31) Additional Resources:

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Epistemic status: hot takes. I have thought quite a bit about EA community building strategy over the past two years, but the style of this post is not "well reasoned and explained". I have been planning to publish something like this for months, kept pushing it back, and now I just want to get it out.

Why I am writing this

For two years now I have been a full-time community builder at EA Switzerland. In those two years I have seen very little high level public discussion about EA Community Building (EA CB) strategy. With this post I hope to spark that discussion and to add my two cents.

With "EA CB" I mean what national or city groups like EA Switzerland do, or should do.

TL;DR

General takes

• "EA community building" is a fuzzy term. It is not clear what it actually means: which programmes count as "general EA CB" and which are "something else".
• EA CB is an unattractive career path. Low pay, yearly grants, low social acceptance outside the bubble.
• It is hard to get funding for EA CB work.
• If the third wave of philanthropy is coming and "the ecosystem" is actually excited [...]

---

Outline:

(00:33) Why I am writing this

(01:01) TL;DR

(03:52) Overview of the field and the current situation

(06:24) Assumptions

(06:52) 1. Positive impact: it does not matter whether the hours are worked by "an EA"

(07:57) 2. Scouting, not converting

(10:12) 3. Cause-area-first outreach produces career changes far more efficiently than principles-first

(12:01) 4. The community people need to boost their productivity is small

(15:01) 5. Once a field is mature, coordination across cause areas has limited value

(18:18) 6. Way more people apply for jobs than for funding

(19:01) 7. If you don't look into what should exist in your region, nobody will

(19:28) What EA CB orgs (regional meta EA orgs) should do at the moment

(20:13) 2. Provide a productive environment, aka a co-working space

(22:56) 3. Build a "captains school"

(23:47) 1. What incubating new projects looks like

(27:57) Doubts

(29:26) Summary: what I want to see less of, and what I want to see more of

(31:52) Thanks

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

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

The swarm has made me feel AI risk in my bones for the first time

Until the Hugging Face / OpenAI swarm thing, I think I deep down felt pretty skeptical that extinction or serious loss of control scenarios were plausible in the next few years.

I’ve been lurking on LessWrong for years, and I remember being freaked out by Yudkowsky's ‘List of Lethalities’ in 2022, and thinking that this did seem intellectually convincing, and maybe I should do something about it. But it didn’t feel that real. And I didn’t do that much about it.

And then I got more involved with EA and AI Safety, and I read more and more, and I found myself feeling more convinced by the more measured, moderate approach to AI risk - AI is going to be transformative, but in a “it’ll destabilise the economy / mass job loss / mass surveillance“ kind of way, and I felt increasingly unimpressed by the OG Yudkowsky/MIRI worldview that we were doomed.

I was distinctly underwhelmed by If Anyone Builds It, Everyone Dies, and in the process of us making a video on it, I began to find it increasingly silly.

[...]

---

Outline:

(00:09) The swarm has made me feel AI risk in my bones for the first time

(02:49) AI is real

(04:19) Progress really isn't slowing down

(04:47) The current models are not aligned by default

(06:17) The companies are really really bad

(07:38) Society is still mostly not waking up to any of this

(09:24) So where does that leave me?

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

Crossposted from my Substack.

~

Suppose the President summons the AI CEOs and his top national security advisors to an emergency meeting at the White House.

He has become extremely concerned about superintelligence — the possibility that AIs far smarter than humanity combined slip beyond our ability to correct or shut down. If that happens, there is no way back. The President is concerned humanity could become permanently out of the driver's seat of its own future. He wants to figure out what to do.

The reaction is panic, chaos, confusion.

The President asks questions. The AI companies are blazing toward superintelligence at high speed — can we slow down as we approach the dangerous thresholds? …Some of the AI companies say they don’t have a good plan to slow down or stop, especially as their competitors may just undercut them if they do. What's that about?

What's going on with China — can we get them to pace as well? Can we get a deal without Beijing sneakily catching up and maybe surpassing us? And if there's no deal to be had, what then?

More like the Cuban Missile Crisis than the NPT

I sometimes hear people [...]

---

Outline:

(01:21) More like the Cuban Missile Crisis than the NPT

(03:24) A scramble and then three phases

(05:18) The scramble: What questions does the President ask?

(09:39) The mechanics of Phase 1

(12:37) A lot of verification work right now is focused on the wrong things

(14:49) What ought we do?

(17:36) Getting to a good scramble

(18:20) Footnotes

The original text contained 1 footnote which was 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.

Headline finding: I audited 17 AI Safety Talent programmes. Zero of 17 have published any comparison group, rejected-applicant follow-up, matched control or randomisation. Not one. Every programme that mentions a counterfactual does it by asking participants to self-report.

Background

At least 70 million dollars has gone into AI safety talent programmes so far, but we still can’t really say much about their impact. Notably, Kairos has just raised 50 million dollars on a self-described "shallow retrospective" that looked at "around 80 people". I decided to do a deep dive into how AI Safety Talent programmes self-evaluate their impact.

Disclaimer, these programmes are small, fast, and generally run by people with no evaluation training and no slack. The oldest one was only established in 2018 (AI Safety Camp, as far as I can tell). Most last less than 6 months. When timelines are genuinely short and urgent, then time spent evaluating is time that could be spent building. I am not criticising any individual programme.

Having said that, there are retrospective evaluation designs that could be run cheaply without slowing anything down. And a field that funds talent pipelines at this scale without knowing whether they work is not moving [...]

---

Outline:

(00:34) Background

(01:52) The audit

(04:52) Thoughts on the counterfactual impact of talent programmes

(08:37) Proposed design

(12:04) Conclusion

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

---

First published:

Source:

---

Narrated by TYPE III AUDIO.

This post was originally published on my Substack. I’m sharing it here because parasite–host interactions in wild fish may have substantial and largely neglected welfare implications, while we still know remarkably little about the experiences of the animals involved. The post is intended mainly to highlight this empirical gap and some potentially tractable directions for further research.

Introduction

Cataracts cover much of both eyes of a fish who is still feeding and moving normally. A bird approaches from above, unseen. The attack succeeds. If anyone is keeping count, the fish is recorded as a death from predation.

This is not a hypothetical case. In fish from Kalmar Sound, the trematode Diplostomum, whose larvae settle in the lens of the eye, was found in 86 of the 88 fish examined, while all 88 had detectable eye-fluke-induced cataracts. Cataracts covered more than half of both eyes in 35% of the sample. Among those tested against a simulated attack from above, none with more than half of both eyes covered responded by fleeing. This does not prove that they were caught more often in the wild [...]

---

Outline:

(00:49) Introduction

(02:59) Weishan: even the census fails

(04:56) Ligula: presence alone is not enough

(06:22) Anisakids: parasites we study for another reason

(08:39) What if the parasites suffer too?

(11:21) How many individuals does this happen to?

(14:16) The numbers do not decide on their own

(15:38) My bet

---

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.