I spend a lot of time narrating various bits of EA/longtermist writing.

The resulting audio exists in many different places. Surprisingly often, people who really like one thing don't know about the other things. This seems bad.[1]

A few people have requested a feed to aggregate 'all Solenoid's narrations.'

Here it is. (Give it a few days to be up on the big platforms.) I'll update it ~weekly.[2]

And here's a list of things I've made or am working on, shared in the hope that more people will discover more things they like:

Human Narrations

• Astral Codex Ten Podcast
  • ~920 episodes so far including all non-paywalled ACX posts and SSC archives going back to 2017, with some classic posts from earlier. 
  • Archive. Patreon.
• LessWrong Curated Podcast 
  • Human narrations of all the Curated posts. Patreon.
• AI Safety Fundamentals 
  • Narrations of most of the [...]

---

Outline:

(00:44) Human Narrations

(02:21) AI Narrations

(02:52) Other things that are like my thing

(03:16) The Future

The original text contained 3 footnotes which were omitted from this narration.

---

First published:

Source:

Narrated by TYPE III AUDIO.

In 1957, Nobel laureate microbiologist Joshua Lederberg and biostatician J. B. S. Haldane sat down together imagined what would happened if the USSR decided to explode a nuclear weapon on the moon. (Maybe they read a recent United Press article about the rumored USSR plans.)

The Cold War was on, Sputnik had recently been launched, and the 40th anniversary of the Bolshevik Revolution was coming up – a good time for an awe-inspiring political statement. Nuking the moon would make a powerful political statement on earth, but the radiation and disruption could permanently harm scientific research on the moon.

What Lederberg and Haldane did not know was that they were onto something – by the next year, the USSR really investigated the possibility of dropping a nuke on the moon. They called it “Project E-4,” one of a series of possible lunar missions.

What [...]

---

Outline:

(01:31) “Why would anyone think it was a good idea to nuke the moon?”

(01:54) Reasons to nuke the moon

(04:33) Reasons to not nuke the moon

(07:00) Afterwards

---

First published:

Source:

Linkpost URL:
https://eukaryotewritesblog.com/2024/04/12/nuking-the-moon/

Narrated by TYPE III AUDIO.

The MIRI Newsletter is back in action after a hiatus since July 2022. To recap some of the biggest MIRI developments since then:

• MIRI released its 2024 Mission and Strategy Update, announcing a major shift in focus: While we’re continuing to support various technical research programs at MIRI, our new top priority is broad public communication and policy change.
  • In short, we’ve become increasingly pessimistic that humanity will be able to solve the alignment problem in time, while we’ve become more hopeful (relatively speaking) about the prospect of intergovernmental agreements to hit the brakes on frontier AI development for a very long time—long enough for the world to find some realistic path forward.
• Coinciding with this strategy change, Malo Bourgon transitioned from MIRI COO to CEO, and Nate Soares transitioned from CEO to President. We also made two new senior staff hires [...]

---

First published:

Source:

Linkpost URL:
https://intelligence.org/2024/04/12/april-2024-newsletter/

Narrated by TYPE III AUDIO.

Time to introduce some more concepts. If an observation is "any data you can receive which affects your actions", then there seem to be two sorts of observations. A plannable observation is the sort of observation where you could plan ahead of time how to react to it. A unplanned observation is the sort which you can't (or didn't) write a lookup-table style policy for.

Put another way, if a policy tells you how to map histories of observations to actions, those "histories" are the plannables. However, to select that policy in the first place, over its competitors, you probably had to do some big computation to find some numbers like "expected utility if I prepare a sandwich when I'm in the kitchen but not hungry", or "the influence of my decisions in times of war on the probability of war in the first place", or "the probability [...]

---

First published:

Source:

Narrated by TYPE III AUDIO.

This article contains more than 100 uses of logical or mathematical notation, so an audio narration would be too hard to follow. You'll find a link to the original text in the episode description.

---

First published:

Source:

Narrated by TYPE III AUDIO.

Below is some advice on making D&D.Sci scenarios. I’m mostly yelling it in my own ear, and you shouldn’t take any of it as gospel; but if you want some guidance on how to run your first game, you may find it helpful.

1. The scoring function should be fair, transparent, and monotonic

D&D.Sci players should frequently be confused, but about how to best reach their goals, not the goals themselves. By the end of the challenge, it should be obvious who won[1].

2. The scoring function should be platform-agnostic, and futureproof

Where possible, someone looking through old D&D.Sci games should be able to play them, and easily confirm their performance after-the-fact. As far as I know, the best way to do this for most challenges is with a HTML/JS web interactive, hosted on github.

3. The challenge should resist pure ML

It should not be possible to reach [...]

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

---

First published:

Source:

Narrated by TYPE III AUDIO.

Atlas Computing is a new nonprofit working to collaboratively advance AI capabilities that are asymmetrically risk-reducing. Our work consists of building prototypes and building an ecosystem around @davidad's Safeguarded AI (formerly referred to as the Open Agency Architecture).

We formed in Oct 2023, and raised nearly $1M, primarily from the Survival and Flourishing Fund and Protocol Labs. We have no physical office, and are currently only Evan Miyazono (CEO) and Daniel Windham (software lead), but over the coming months and years, we hope to create compelling evidence that:

1. The Safeguarded AI research agenda includes both research and engineering projects where breakthroughs or tools can incrementally reduce AI risks.
2. If Atlas Computing makes only partial progress toward building safeguarded AI, we’ll likely have put tools into the world that are useful for accelerating human oversight and review of AI outputs, asymmetrically favoring risk reduction.
3. When davidad's ARIA program concludes [...]

---

Outline:

(01:18) Our overall strategy

(02:18) Specification-based AI vs other AI risk mitigation strategies

(03:26) How this differs from davidad's work at ARIA

(05:02) Our plan

(06:34) Why a nonprofit?

(07:31) How to engage with us

---

First published:

Source:

Narrated by TYPE III AUDIO.

This was originally posted on Nathaniel's and Nuno's substacks (Pending Survival and Forecasting Newsletter, respectively). Subscribe here and here!

Discussion is also occurring on the EA Forum here (couldn't link the posts properly for technical reasons).

When the Effective Altruism, Bay Area rationality, judgemental forecasting, and prediction markets communities think about risk, they typically do so along rather idiosyncratic and limited lines. These overlook relevant insights and practices from related expert communities, including the fields of disaster risk reduction, safety science, risk analysis, science and technology studies—like the sociology of risk—and futures studies.

To remedy this state of affairs, this document—written by Nathaniel Cooke and edited by Nuño Sempere—(1) explains how disaster risks are conceptualised by risk scholars, (2) outlines Normal Accident Theory and introduces the concept of high-reliability organisations, (3) summarises the differences between “sexy” and “unsexy” global catastrophic risk (GCR) scenarios, and (4) [...]

---

Outline:

(02:00) 1. Disaster Risk Models

(03:21) 1.1 The Determinants of Risk

(09:34) 1.2 Exposure

(10:31) 1.3 Pressure and Release

(12:35) 1.4 The Disaster Cycle

(14:39) 1.5 Complexity and Causation

(18:01) 2 Normal Accidents and High-Reliability Organisations

(23:30) 3 Sexy and Unsexy Risks

(26:11) 4 Forecasting, Foresight, and Futures

(28:55) 4.1 Forecasting

(34:01) 4.2 Foresight

(39:12) 4.3 Futures Studies

(42:21) Conclusion

The original text contained 10 footnotes which were omitted from this narration.

---

First published:

Source:

Narrated by TYPE III AUDIO.

Youtube Video

Recently, I was interviewed by Henry Sleight and Mikita Balesni about how I select alignment research projects. Below is the slightly cleaned up transcript for the YouTube video.

Introductions

Henry Sleight: How about you two introduce yourselves?

Ethan Perez: I'm Ethan. I'm a researcher at Anthropic and do a lot of external collaborations with other people, via the Astra Fellowship and SERI MATS. Currently my team is working on adversarial robustness, and we recently did the sleeper agents paper. So, basically looking at we can use RLHF or adversarial training or current state-of-the-art alignment safety training techniques to train away bad behavior. And we found that in some cases, the answer is no: that they don't train away hidden goals or backdoor behavior and models. That was a lot of my focus in the past [...]

---

Outline:

(00:14) Youtube Video

(00:29) Introductions

(02:04) How Ethan Selects Research Projects

(02:43) Top-down vs Bottom-up Approach

(05:48) Empirical Feedback and Prototyping

(08:01) Sharing Ideas and Getting Feedback

(08:43) Pivoting Projects Based on Promising Results

(10:08) Top-down Approach and Low-Hanging Fruit

(13:32) Lessons Learned when Choosing Research Projects

(13:45) Reading Prior Work

(17:19) Duplication of Work and Collaborations

(18:24) Deciding to Collaborate or Not

(21:05) Advice for Junior Researchers

(23:45) Pivoting projects

(29:13) Red Flags for Projects

(31:16) Open Source Infrastructure Needs

(33:41) Tracking Multiple Promising Research Directions and Switching

(37:31) Visibility into Alternatives

(39:00) Final Takeaway

---

First published:

Source:

Narrated by TYPE III AUDIO.

This is a followup to the D&D.Sci post I made ten days ago; if you haven’t already read it, you should do so now before spoiling yourself.

Here is the web interactive I built to let you evaluate your solution; below is an explanation of the rules used to generate the dataset (my full generation code is available here, in case you’re curious about details I omitted). You’ll probably want to test your answer before reading any further.

Ruleset

Turtle Types

There are three types of turtle present in the swamp: normal turtles, clone turtles, and vampire turtles.

Clone turtles are magically-constructed beasts who are mostly identical. They always have six shell segments, bizarrely consistent physiology, and a weight of exactly 20.4lb. Harold is a clone turtle.

Vampire turtles can be identified by their gray skin and fangs. They’re mostly like regular turtles, but their flesh no longer obeys [...]

---

Outline:

(00:37) Ruleset

(00:40) Turtle Types

(01:20) Turtle characteristics

(01:24) Age

(01:47) Wrinkles

(01:55) Scars

(02:04) Shell Segments

(02:17) Color

(02:31) Miscellaneous Abnormalities

(03:02) Nostril Size

(03:13) Turtle Weight

(03:30) Flesh Weight

(03:41) Shell Weight

(03:58) Mutation Weight

(04:17) Strategy

(04:26) Leaderboard

(04:31) Reflections

(05:51) Scheduling

The original text contained 3 footnotes which were omitted from this narration.

---

First published:

Source:

Narrated by TYPE III AUDIO.