Bloomberg reports that OpenAI internally has benchmarks for “Human-Level AI.” They have 5 levels, with the first being the achieved level of having intelligent conversation, to level 2, “[unassisted, PhD-level] Reasoners,” level 3, “Agents,” level 4, systems that can “come up with new innovations,” and finally level 5, “AI that can do the work of… Organizations.”

The levels, in brief, are:

1 - Conversation
2 - Reasoning
3 - Agent
4 - Innovation
5 - Organization

This is being reported secondhand, but given that, there seem to be some major issues with the ideas. Below, I outline two major issues I have with what is being reported.

...but this is Superintelligence 

First, given the levels of capability being discussed, OpenAI's typology is, at least at higher levels, explicitly discussing superintelligence, rather than “Human-Level AI.” To see this, I’ll use Bostrom's admittedly imperfect definitions. He starts by [...]

---

Outline:

(00:58) ...but this is Superintelligence

(03:00) Ordering Levels?

---

First published:

July 15th, 2024

Source:

https://www.lesswrong.com/posts/JycQ8dZwAKcgfbWig/misnaming-and-other-issues-with-openai-s-human-level

---

Narrated by TYPE III AUDIO.

This is a link post.

A few days ago, Gray Swan published code and models for their recent “circuit breakers” method for language models.[1]1

The circuit breakers method defends against jailbreaks by training the model to erase “bad” internal representations. We are very excited about data-efficient defensive methods like this, especially those which use interpretability concepts or tools.

At the link, we briefly investigate three topics:

1. Increased refusal rates on harmless prompts: Do circuit breakers really maintain language model utility? Most defensive methods come with a cost. We check the model's effectiveness on harmless prompts, and find that the refusal rate increases from 4% to 38.5% on or-bench-80k.
2. Moderate vulnerability to different token-forcing sequences: How specialized is the circuit breaker defense to the specific adversarial attacks they studied? All the attack methods evaluated in the circuit breaker paper rely on a “token-forcing” optimization objective which maximizes the likelihood of [...]

---

First published:

July 14th, 2024

Source:

https://www.lesswrong.com/posts/NAYyHimM3FaDYLvEH/breaking-circuit-breakers

---

Narrated by TYPE III AUDIO.

Glue your colored whiteboard makers together with duck tape to create a whiteboard pen magazine. If required glue a paper layer around the duck tape to make it non-sticky to the touch. Mark which side is up with an arrow, such that you can always in the same way.

This more than doubled my color-switching speed, fits in my pocket, and takes <5m to make.

It lasts: When a pen is empty just insert a new one. No need to create another magazine. No signs of degradation after multiple hundred unholstering-holstering cycles in total.

FUUU 754 Standard: I am using RFC2119. The magazine MUST be vertically stacked (like shown in the pictures). The magazine SCHOULD have a visual mark, that is clearly visible when the magazine lays on a flat surface, to indicate the top side of the magazine (like shown in the first [...]

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

---

First published:

July 12th, 2024

Source:

https://www.lesswrong.com/posts/eBo5YCHP5H58Y5Qjx/whiteboard-pen-magazines-are-useful

---

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.

It's quite possible someone has already argued this, but I thought I should share just in case not.

Goal-Optimisers and Planner-Simulators

When people in the past discussed worries about AI development, this was often about AI agents - AIs that had goals they were attempting to achieve, objective functions they were trying to maximise. At the beginning we would make fairly low-intelligence agents, which were not very good at achieving things, and then over time we would make them more and more intelligent. At some point around human-level they would start to take-off, because humans are approximately intelligent enough to self-improve, and this would be much easier in silicon.

This does not seem to be exactly how things have turned out. We have AIs that are much better than humans at many things, such that if a human had these skills we would think they were extremely capable. And [...]

---

Outline:

(00:10) Goal-Optimisers and Planner-Simulators

(01:17) What is the significance for existential risk?

(02:43) How could we react to this?

---

First published:

July 14th, 2024

Source:

https://www.lesswrong.com/posts/yRqzRmEC3aqcajKjy/llms-as-a-planning-overhang

---

Narrated by TYPE III AUDIO.

This is a link post.

I argue here that preventing a large iceberg from melting is absurdly cheap per unit area compared to just about any other way of making new land, and it's kind of crazy to spend money on space exploration and colonization before colonizing the oceans with floating ice-islands.

---

First published:

July 13th, 2024

Source:

https://www.lesswrong.com/posts/gthjxPDywrMTs3p2j/ice-the-penultimate-frontier

---

Narrated by TYPE III AUDIO.

Alex Turner introduced an exercise to test subjects’ ability to notice falsehoods: change factual statements in Wikipedia articles, hand the edited articles to subjects and see whether they notice the modifications.

I’ve spent a few hours making such modifications and testing the articles on my friend group. You can find the articles here. I describe my observations and thoughts below. The bottom line: it is hard to come up with good modifications / articles to modify, and this is the biggest crux for me.

The concept

Alex Turner explains the idea well here. The post is short, so I'm just copying it here:

Rationality exercise: Take a set of Wikipedia articles on topics which trainees are somewhat familiar with, and then randomly select a small number of claims to negate (negating the immediate context as well, so that you can't just syntactically discover which claims were negated).

For [...]

---

Outline:

(00:38) The concept

(02:14) Features of good modifications

(04:44) Examples

(05:40) Takeaways and thoughts

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

---

First published:

July 14th, 2024

Source:

https://www.lesswrong.com/posts/4fqdfrDzFebsx8amf/brief-notes-on-the-wikipedia-game

---

Narrated by TYPE III AUDIO.

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

This post is the version of Yudkowsky's argument for inner misalignment that I wish I'd had in my head a few years ago. I don't claim that it's novel, that I endorse it, or even that Yudkowsky would endorse it; it's primarily an attempt to map his ideas into an ontology that makes sense to me (and hopefully others).

This post is formulated in terms of three premises, which I explore in turn. My arguments deliberately gloss over some nuances and possible objections; in a follow-up post, I'll explore three of them. In a third post I'll dig into the objection I find most compelling, and outline a research agenda that aims to flesh it out into a paradigm for thinking about cognition more generally, which I'm calling coalitional agency.

Background

An early thought experiment illustrating the [...]

---

Outline:

(00:53) Background

(02:22) Increasing intelligence requires compressing representations

(04:35) The simplest goals are highly decomposable broadly-scoped utility functions

(08:50) Increasingly intelligent AIs will converge towards squiggle-maximization

---

First published:

July 13th, 2024

Source:

https://www.lesswrong.com/posts/fjfWrKhEawwBGCTGs/a-simple-case-for-extreme-inner-misalignment

---

Narrated by TYPE III AUDIO.

This is a link post.

This linkpost contains a lightly-edited transcript of highlights of my recent AI x-risk debate with Robin Hanson, and a written version of what I said in the post-debate analysis episode of my Doom Debates podcast.

Introduction

I've poured over my recent 2-hour AI x-risk debate with Robin Hanson to clip the highlights and write up a post-debate analysis, including new arguments I thought of after the debate was over.

I've read everybody's feedback on YouTube and Twitter, and the consensus seems to be that it was a good debate. There were many topics brought up that were kind of deep cuts into stuff that Robin says.

On the critical side, people were saying that it came off more like an interview than a debate. I asked Robin a lot of questions about how he sees the world and I didn't "nail" him. And people were [...]

---

Outline:

(00:26) Introduction

(06:47) Robins AI Timelines

(09:46) Culture vs. Intelligence

(13:23) Innovation Accumulation vs. Intelligence

(16:13) Optimization = Collecting + Spreading?

(17:28) Worldwide Growth

(20:29) Extrapolating Robust Trends

(22:55) Seeing Optimization-Work

(24:58) Exponential Growth With Respect To...

(27:34) Can Robins Methodology Notice Foom In Time?

(38:06) Foom Argument As Conjunction

(41:12) Headroom Above Human Intelligence

(46:37) More on Culture vs. Intelligence

(51:42) The Goal-Completeness Dimension

(57:12) AI Doom Scenario

(01:02:36) Corporations as Superintelligences

(01:06:18) Monitoring for Signs of Superintelligence

(01:08:07) Will AI Societys Laws Protect Humans?

(01:11:58) Feasibility of ASI Alignment

(01:14:34) Robins Warning Shot

(01:16:10) The Cruxes

(01:17:06) How to Present the AI Doom Argument

(01:19:13) About Doom Debates

The original text contained 3 images which were described by AI.

---

First published:

July 12th, 2024

Source:

https://www.lesswrong.com/posts/quj3RniAw7GGJtqeX/robin-hanson-ai-x-risk-debate-highlights-and-analysis

---

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.

Epistemic status: Mostly speculation and anecdotes, feel free to skip to The List once you understand its reason for existence.

tldr: This is a curated list of observable details about a person's appearance that indicate something useful/surprising about them. Ideally, studying this list will be an efficient way to cultivate more insightful observational/abductive abilities, approaching the fictional example of Sherlock Holmes. Please contribute in the comments section after reading the Rules.

Background

Is it possible to develop observational abilities comparable to Sherlock Holmes?

Sir Arthur Conan Doyle's fictional detective has many enviable skills, including mastery of disguise and some expertise at unarmed combat, as well as generally being a genius, but we will focus primarily on his more well known observational power. Though Holmes is often described as a master of logical "deduction," this power is better described as (possibly superhuman) abduction. That is, Holmes perceives tiny details that many [...]

---

Outline:

(00:35) Background

(05:34) Rules

(06:50) The List

(06:52) Cauliflower Ear - Combat Sports

---

First published:

December 2nd, 2023

Source:

https://www.lesswrong.com/posts/Yz33koDN5uhSEaB6c/sherlockian-abduction-master-list

---

Narrated by TYPE III AUDIO.

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

TLDR: We’re hiring two research assistants to work on advancing developmental interpretability and other applications of singular learning theory to alignment.

About Us

Timaeus's mission is to empower humanity by making breakthrough scientific progress on alignment. Our research focuses on applications of singular learning theory to foundational problems within alignment, such as interpretability (via “developmental interpretability”), out-of-distribution generalization (via “structural generalization”), and inductive biases (via “geometry of program synthesis”). Our team spans Melbourne, the Bay Area, London, and Amsterdam, collaborating remotely to tackle some of the most pressing challenges in AI safety.

For more information on our research and the position, see our Manifund application, this update from a few months ago, our previous hiring call and this advice for applicants.

Position Details

• Title: Research Assistant
• Location: Remote
• Duration: 6-month contract with potential for extension.

• ---

  Outline:

  (00:22) About Us

  (01:16) Position Details

  (01:46) Key Responsibilities

  (02:20) Projects

  (03:30) Qualifications

  (04:21) Application Process

  (04:45) Apply Now

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

  ---

First published:

July 12th, 2024

Source:

https://www.lesswrong.com/posts/w7WdcKJgcwkGAzuse/timaeus-is-hiring

---

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.