LessWrong posts by zvi

“AI #132 Part 2: Actively Making It Worse” by Zvi

Listen

Description

It's rough out there. Have we tried engaging in less active sabotage? No? Carry on.

Table of Contents

  1. Quiet Speculations. What will become the new differentiators?
  2. The Quest for Sane Regulations. Bostrom proposes improving on status quo a bit.
  3. The Quest For No Regulations. Cato Institute CEO says Cato Institute things.
  4. But This Time You’ve Gone Too Far. You’re drawing the line where? Really?
  5. Chip City. Sabotaging American solar and wind, the strategic value of chips.
  6. The Week in Audio. Interest rates, Lee versus Piper, Jack Clark, Hinton.
  7. Rhetorical Innovation. Listening does not accomplish what you might hope.
  8. Safety Third at xAI. More on their no good very bad framework. A new prompt.

  9. Misaligned! Will any old crap cause misalignment? At least a little, yes.
  10. Lab Safeguards Seem Inadequate. AI Safety Claims formalizes how inadequate.
  11. [...]

---

Outline:

(00:20) Quiet Speculations

(00:53) The Quest for Sane Regulations

(04:34) The Quest For No Regulations

(07:36) But This Time You've Gone Too Far

(16:05) Chip City

(25:13) The Week in Audio

(27:25) Rhetorical Innovation

(36:19) Safety Third at xAI

(44:48) Misaligned!

(46:03) Lab Safeguards Seem Inadequate

(47:52) Aligning a Smarter Than Human Intelligence is Difficult

(54:37) The Lighter Side

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

---

First published:

September 5th, 2025


Source:

https://www.lesswrong.com/posts/jppFpbRCG9y3Xyuau/ai-132-part-2-actively-making-it-worse

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Screenshot showing safety instructions for resisting
A promotional image pairing a T-800 Terminator with the director.
Browserbase identity card for travel agent with future date 2025.
A diagram showing
Two graphs comparing Huawei and Nvidia chip manufacturing and performance metrics through 2026.</p><p>The top graph shows manufacturing volume projections, while the bottom scatter plot compares chip performance parameters across different models and timelines. The visualization clearly demonstrates Nvidia's significant lead in both manufacturing scale and technical capabilities.
These appear to be safety instructions and guidelines for content moderation. The text outlines key principles for handling queries, including rules about disallowed activities, appropriate response levels, and content policies.</p><p>The main points include:<br /> - Instructions for handling different types of queries<br /> - Guidelines for assuming good intent<br /> - Rules about factual responses<br /> - Content policy parameters</p><p>The text is displayed in white on a dark background.
Industrial AGI machine with Tool AI boxes in a pastoral countryside setting.</p><p>The image shows a futuristic device with a glowing green
Chart showing AI company threat preparedness across five major organizations, comparing security metrics.</p><p>The image displays a color-coded matrix evaluating three risk categories: Security, Misalignment risk prevention, and Misuse prevention. The companies are represented by logos at the top, and performance is rated from
Miles Brundage tweets:
Yohei tweets:
Agnes Callard tweets:
Alexander Horner tweets:
Sawyer Hood tweets:
Secretary Chris Wright tweets:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.