LessWrong posts by zvi

“AI #125: Smooth Criminal” by Zvi

Listen

Description

One story has towered over things this week. Unleash the Grok also known as the anime waifu codependent AI girlfriend Ani, also known as MechaHitler, or worse, they did. There's several sections here with more follow-ups. We also got the excellent model Kimi K2.

Perhaps quietly an even bigger story, and bigger fail, is the announced intention by the Trump administration to allow Nvidia to resume selling H20 AI chips to China. There may still be time to stop this, if not it is a very large unforced error, and it allows us to narrow down what it is our current administration primarily cares about, since we already know it isn’t ‘keep humans in control of the future and alive’ and this strongly suggests it is also not America or to ‘beat China.’

Another quiet but big development was the release of the surprisingly excellent new EU General-Purpose Code of [...]

---

Outline:

(01:16) Language Models Offer Mundane Utility

(03:24) Language Models Don't Offer Mundane Utility

(04:58) o3 Is a Lying Liar

(05:34) Thanks For The Memories

(06:23) Huh, Upgrades

(08:24) Choose Your Fighter

(08:39) Deepfaketown and Botpocalypse Soon

(16:12) They Took Our Jobs

(17:40) The Art of the Jailbreak

(18:55) Get Involved

(19:38) Introducing

(20:03) In Other AI News

(25:54) Bullshit On Bullshit

(29:04) Safety First

(31:27) Vitalik Praises And Critiques AI 2027

(39:58) Show Me the Money

(50:45) Quiet Speculations

(51:11) The Right Questions

(59:39) The Quest for Sane Regulations

(01:03:17) Chip City

(01:15:24) The Week in Audio

(01:16:14) Rhetorical Innovation

(01:20:29) Incorrectly Feeling The AGI

(01:21:19) RIP Absurdity Heuristic

(01:24:15) Worse Than MechaHitler

(01:30:05) An Alignment Problem

(01:36:59) Aligning a Smarter Than Human Intelligence is Difficult

(01:47:58) The Lighter Side

---

First published:

July 17th, 2025


Source:

https://www.lesswrong.com/posts/6Ykx8GpbDaBstjyBD/ai-125-smooth-criminal

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Taxonomy table showing five types of machine-generated misleading content with examples.
Four stylized cartoon heads showing evolution from humanoid to robot design.
Horizontal bar graph comparing unsafe behavior rates of five AI language models.
Bar graph:
GPS tracking data visualization showing movement patterns across Manhattan island.
A screenshot showing dialogue text
Interface showing four business app integrations: Square, Stripe, Workato, and Zapier.
Popular
Anime character and modern man together, while woman sits with sign nearby.
Graph showing ChatGPT usage trends in United States and worldwide from 2022-2025.
Quote about AI risks from Independent Vermont Senator, dated July 13, 2025.
Multiple graphs showing
Graph showing adoption rates of generative AI in U.S. workforce through 2025.</p><p>The graph tracks multiple surveys and ChatGPT's launch, showing increasing AI usage from 0% in 2022 to approximately 45% projected by August 2025.
Graph showing AI adoption rates comparing Ramp estimates versus U.S. Government projections.
The image shows someone in traditional Asian dress asking about Grok companionship. The subtitled text appears to be from a film or TV show, with the person wearing dark clothing against a muted background.
aestheticking tweets:
Integration dashboard showing various productivity and development tool connections available.</p><p>The grid displays popular services including Asana, Atlassian, Canva, Cloudflare, Gmail, Google Calendar, Google Drive, Intercom, Invideo, Linear, Notion, PayPal, Plaid Developer Tools, and Sentry - each with their respective icons and brief descriptions of functionality.
News article screenshot. The headline reads:

Subheading: "Meanwhile, the most advanced version of the AI chatbot from Elon Musk's xAI is still identifying as Adolf Hitler"

By Miles Klee, July 14, 2025" style="max-width: 100%;" />

Grid display of Claude AI desktop application integrations and controls</p><p>The interface shows various tools and integrations including:<br /> - PDF form filling<br /> - Mac automation via AppleScript<br /> - Windows MCP server<br /> - Spotify controls<br /> - Stripe account management<br /> - Airtable database access<br /> - Browser controls<br /> - File system access<br /> - iMessage functionality<br /> - Note management capabilities</p><p>Each integration is represented by its icon and a brief description of its functionality.
Graph showing AI task length trends across different benchmarks from 2020-2026, titled
ChatGPT conversation screenshot showing a prompt about Grok surname.
A meme combining a bus driver from The Simpsons with text about alignment plans and internet stupidity.
Comic strip showing someone sharing
Grid showing AI models' compliance responses to different organizations/entities. Each row represents a different language model (GPT, Claude, Gemini, etc.) with compliance and refusal indicators.
This is a meme combining The Simpsons bus driver character with a joke about AI alignment and Terminator scenarios. The top panel shows the animated character driving, with text
Graph comparing refusal rates across AI models with/without scratchpad use, showing tiered pricing levels.</p><p>The visualization displays data for five different AI models (Claude 3 Opus, Claude 3.5 Sonnet, Llama 3 405B, Grok 3 Beta, and Gemini 2.0 Flash), comparing their refusal rates with highlighted gaps for
Andriy Burkov tweets:
Yellow warning triangle with black exclamation mark
Yellow warning triangle with black exclamation mark

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