Developers of frontier AI systems will face increasingly challenging decisions about whether their AI systems are safe enough to develop and deploy. One reason why systems may not be safe is if they engage in scheming. In our new report "Towards evaluations-based safety cases for AI scheming", written in collaboration with researchers from the UK AI Safety Institute, METR, Redwood Research and UC Berkeley, we sketch how developers of AI systems could make a structured rationale – 'a safety case' – that an AI system is unlikely to cause catastrophic outcomes through scheming.

Note: This is a small step in advancing the discussion. We think it currently lacks crucial details that would be required to make a strong safety case.

Read the full report.

Figure 1. A condensed version of an example safety case sketch, included in the report. Provided for illustration.

Scheming and Safety Cases

For [...]

---

Outline:

(01:13) Scheming and Safety Cases

(02:01) Core Arguments and Challenges

(03:42) Safety cases in the near future

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

The original text contained 1 image which was described by AI.

---

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.

We (Connor Leahy, Gabriel Alfour, Chris Scammell, Andrea Miotti, Adam Shimi) have just published The Compendium, which brings together in a single place the most important arguments that drive our models of the AGI race, and what we need to do to avoid catastrophe.

We felt that something like this has been missing from the AI conversation. Most of these points have been shared before, but a “comprehensive worldview” doc has been missing. We’ve tried our best to fill this gap, and welcome feedback and debate about the arguments. The Compendium is a living document, and we’ll keep updating it as we learn more and change our minds.

We would appreciate your feedback, whether or not you agree with us:

• If you do agree with us, please point out where you think the arguments can be made stronger, and contact us if there are [...]

---

First published:

Source:

Narrated by TYPE III AUDIO.

There are two nuclear options for treating depression: Ketamine and TMS; This post is about the latter.

TMS stands for Transcranial Magnetic Stimulation. Basically, it fixes depression via magnets, which is about the second or third most magical things that magnets can do.

I don’t know a whole lot about the neuroscience - this post isn’t about the how or the why. It's from the perspective of a patient, and it's about the what.

What is it like to get TMS?

TMS

The Gatekeeping

For Reasons™, doctors like to gatekeep access to treatments, and TMS is no different. To be eligible, you generally have to have tried multiple antidepressants for several years and had them not work or stop working. Keep in mind that, while safe, most antidepressants involve altering your brain chemistry and do have side effects.

Since TMS is non-invasive, doesn’t involve any drugs, and has basically [...]

---

Outline:

(00:35) TMS

(00:38) The Gatekeeping

(01:49) Motor Threshold Test

(04:08) The Treatment

(04:15) The Schedule

(05:20) The Experience

(07:03) The Sensation

(08:21) Results

(09:06) Conclusion

The original text contained 2 images which were described by AI.

---

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.

In episode 158 of Cortext podcast, CGP Grey gives their high-level reason why they are worried about AI.

My one line summary: AI should not be compared to nuclear weapons but instead to biological weapons or memes, which evolve under the implicit evolutionary pressures that exist, leading to AI's that are good at surviving and replicating.

The perspective is likely known by many in the community already, but I had not heard it before. Interestingly, there have actually been experiments where they just put random strings of code in an environment where they interact, and self-replicating code appeared. See Cognitive Revolution podcast on 'Computational Life: How Self-Replicators Arise from Randomness', with Google researchers Ettore Randazzo and Luca Versari.

I quote the relevant part of the podcast below, but I recommend listening because the emotion and delivery is impactful. It is from 1:22:00 onwards.

To be explicit and not [...]

---

First published:

Source:

Narrated by TYPE III AUDIO.

I hope it has some educational, memetic or at least humorous potential.

---

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.

We’re coming out firmly against it.

Our attitude:

The customer is always right. Yes, you should go ahead and fix your own damn pipes if you know how to do that, and ignore anyone who tries to tell you different. And if you don’t know how to do it, well, it's at your own risk.

With notably rare exceptions, it should be the same for everything else.

I’ve been collecting these for a while. It's time.

Campaign Talk

Harris-Walz platform includes a little occupational licensing reform, as a treat.

Universal Effects and Recognition

Ohio's ‘universal licensing’ law has a big time innovation, which is that work experience outside the state actually exists and can be used to get a license (WSJ).

Occupational licensing decreases the number of Black men in licensed professions by up to 19% [...]

---

Outline:

(00:43) Campaign Talk

(00:52) Universal Effects and Recognition

(03:57) Construction

(04:08) Doctors and Nurses

(05:01) Florists

(07:32) Fortune Telling

(09:41) Hair

(14:23) Lawyers

(16:07) Magicians

(16:36) Military Spouses

(17:21) Mountain Climbing

(18:07) Music

(18:20) Nurses

(19:49) Physical Therapists

(20:09) Whatever Could Be Causing All This Rent Seeking

(21:42) Tornado Relief

(22:10) Pretty Much Everything

The original text contained 9 images which were described by AI.

---

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.

We begin with three stories about three people.

First, Zhu Di, emperor of China from 1402 to 1424. In that period, it was traditional for foreign envoys to present gifts to the emperor and make a show of submission, reinforcing the emperor's authority and China's image as the center of civilization. Yet the emperor would send the envoys off with gifts in return, often worth more than the gifts the envoy had given - suggesting that the emperor's authority and dominance did not actually translate into much bargaining power.

Second, Kevin Systrom, one of the founders of Instagram. When Instagram was bought by Facebook for $1B in 2012, it had only 13 employees. Systrom presumably found himself with a great deal of money, the most direct form of bargaining power amongst humans. Yet with only 13 employees, he commanded little of the obedience or authority displayed by an emperor.

---

Outline:

(01:47) Emperor of China: Dominance

(03:36) Instagram Founder: Bargaining Power

(05:19) Benjamin Jesty: Getting What You Want

---

First published:

Source:

Narrated by TYPE III AUDIO.

7.1 Post summary / Table of contents

This is the 7th of a series of 8 blog posts, which I’m serializing weekly. (Or email or DM me if you want to read the whole thing right now.)

The main thrust of this post is an opinionated discussion of the book The Origin of Consciousness in the Breakdown of the Bicameral Mind by Julian Jaynes (1976) in §7.4, and then some discussion of how hallucinations and delusions arise from schizophrenia and mania (§7.5), and from psychotic depression and BPD (§7.6).

But first—what exactly do hallucinations have to do with intuitive self-models? A whole lot, it turns out!

We’ve seen throughout this series that different states of consciousness can have overlapping mechanisms. Deep trance states (Post 4) mechanistically have a lot in common with Dissociative Identity Disorder (Post 5). Lucid trance states (Post 4) mechanistically have a lot in common with [...]

---

Outline:

(00:07) 7.1 Post summary / Table of contents

(03:40) 7.2 Against an axiomatic bright-line distinction between “hallucinations” and normal inner speech and imagination

(05:45) 7.3 Culture and psychology can both influence intuitive self-models

(07:24) 7.4 Hallucinations via culture

(07:29) 7.4.1 Hearing voices as a cultural norm

(08:55) 7.4.2 Sorry Julian Jaynes, but hearing the voice of a god does not make you an “automaton” under “hallucinatory control”

(11:00) 7.4.2.1 The counterargument from common sense

(11:44) 7.4.2.2 The counterargument from archaeological and historical evidence

(13:07) 7.4.2.3 The counterargument from psychology

(16:33) 7.4.3 “Hallucinogenic aids”

(19:57) 7.4.3.1 Desensitization to hallucinogenic aids as a possible explanation of shifts in culture and iconography around 1000 BC

(23:33) 7.5 Hallucinations and delusions via disjointed cognition (schizophrenia and mania)

(28:05) 7.6 Hallucinations and delusions via strong emotions (BPD and psychotic depression)

(30:22) Conclusion

The original text contained 7 footnotes which were omitted from this narration.

The original text contained 2 images which were described by AI.

---

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.

Recent discussions about artificial intelligence safety have focused heavily on ensuring AI systems remain under human control. While this goal seems laudable on its surface, we should carefully examine whether some proposed safety measures could paradoxically enable rather than prevent dangerous concentrations of power.

The Control Paradox

The fundamental tension lies in how we define "safety." Many current approaches to AI safety focus on making AI systems more controllable and aligned with human values. But this raises a critical question: controllable by whom, and aligned with whose values?

When we develop mechanisms to control AI systems, we are essentially creating tools that could be used by any sufficiently powerful entity - whether that's a government, corporation, or other organization. The very features that make an AI system "safe" in terms of human control could make it a more effective instrument of power consolidation.

Natural Limits on Human Power

Historical [...]

---

Outline:

(00:25) The Control Paradox

(01:05) Natural Limits on Human Power

(02:46) The Human-AI Nexus

(03:58) Alignment as Enabler of Coherent Entities

(04:21) Dynamics of Inevitable Control?

(05:14) The Offensive Advantage

(06:12) The Double Bind of Development

(06:40) Rethinking Our Approach

(08:39) Conclusion

The original text contained 5 footnotes which were omitted from this narration.

---

First published:

Source:

Narrated by TYPE III AUDIO.

This is a follow-up to last week's D&D.Sci scenario: if you intend to play that, and haven't done so yet, you should do so now before spoiling yourself.

There is a web interactive here you can use to test your answer, and generation code available here if you're interested, or you can read on for the ruleset and scores.

RULESET

A character's class/race/level/items boil down to giving them scores in two separate stats: Speed and Power.

Stat Calculation

Each parameter of a character has an effect on their stats:

• Class: Your class provides base stats as follows:
  • Knight: 2 Speed/12 Power
  • Warrior: 4 Speed/10 Power
  • Ranger: 6 Speed/8 Power
  • Monk: 8 Speed/6 Power
  • Fencer: 10 Speed/4 Power
  • Ninja: 12 Speed/2 Power
• Race: Your race modifies power as follows:
  • Dwarf: -3 Speed, +3 Power
  • Human: No Effect
  • Elf: +3 Speed, -3 Power
• Level: Each level provides +1 [...]

---

Outline:

(00:28) RULESET

(00:39) Stat Calculation

(03:03) Combat

(04:57) STRATEGY

(08:29) BONUS OBJECTIVE

(10:28) LEADERBOARD

(10:59) DATASET GENERATION

(11:42) REFLECTIONS and FEEDBACK REQUEST

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

---

First published:

Source:

Narrated by TYPE III AUDIO.