- The Rundown AI
- Posts
- Anthropic maps AI's moral compass
Anthropic maps AI's moral compass
PLUS: New UAE initiative plans to let AI write laws
Rowan Cheung
April 22, 2025
Good morning, AI enthusiasts. Anthropic just pulled back the curtain on AI morality — revealing the first-ever map of Claude's real-world values based on hundreds of thousands of actual conversations.
With AI systems increasingly shaping our decisions (and now even legislation), cracking the alignment code behind models’ moral compass has never been more important.
Reminder: Our next workshop is today at 3 PM EST with Superhuman, where you’ll learn to fully optimize your email system with AI and eliminate inbox chaos. RSVP here.
In today’s AI rundown:
Anthropic charts Claude's values
UAE plans to let AI write the laws
Research with NotebookLM web discovery
Hassabis: AI could end all disease
4 new AI tools & 4 job opportunities
LATEST DEVELOPMENTS
ANTHROPIC
Image source: o3 / The Rundown
The Rundown: Anthropic just published a study analyzing hundreds of thousands of real AI conversations to understand how models like Claude make moral judgements — building the first large‑scale map of the model's values in day‑to‑day interactions.
The details:
Researchers analyzed over 300,000 real (but anonymous) conversations to find and categorize 3,307 unique values expressed by the AI.
They found 5 types of values (Practical, Knowledge-related, Social, Protective, Personal), with Practical and Knowledge-related being the most common.
Values like helpfulness and professionalism appeared most frequently, while ethical values were more common during resistance to harmful requests.
Claude's values also shifted based on context, such as emphasizing "healthy boundaries" in relationship advice vs "human agency" in AI ethics discussions.
Why it matters: AI is increasingly shaping real-world decisions and relationships, making understanding their actual values more crucial than ever. This study also moves the alignment discussion toward more concrete observations, revealing that AI’s morals and values may be more contextual and situational than a static point of view.
TOGETHER WITH VANTA
The Rundown: Join Vanta, Wiz, and Modo Labs on May 8 for a live fireside chat that demystifies how leading companies evolve GRC and cloud defenses — so you can scale security without stalling innovation.
This interactive discussion will include:
Insights into top-of-mind issues for GRC and security pros
Actionable steps to strengthen your existing program by actioning on risks
Key strategies for integrating AI capabilities into your security operations
AI & GOVERNMENT
Image source: Reve / The Rundown
The Rundown: The United Arab Emirates unveiled plans to become the first nation to integrate AI directly into its lawmaking process, establishing a new government unit to oversee the transformation of how laws are written, reviewed, and updated.
The details:
A new Regulatory Intelligence Office will lead the initiative, which aims to cut legislative development time by 70% through AI-assisted drafting and analysis.
The system will use a database combining federal and local laws, court decisions, and government data to suggest legislation and amendments.
The plan builds on the UAE’s major investments in AI, including a dedicated $30B AI-focused infrastructure fund through its MGX investment platform.
The move was met with mixed reactions, with experts warning of the tech’s reliability, bias, and interpretive issues present in training data.
Why it matters: While many governments have already begun integrating AI into their ranks, this is one of the first examples of giving it legislative power in some capacity. As systems reach superhuman levels of persuasion, reasoning, and more, their use in politics will raise existential questions about AI vs. human judgment in lawmaking.
AI TRAINING
The Rundown: In this tutorial, you will learn how to use NotebookLM's new Discover Sources feature to find and add relevant web sources to your notebook with just a few clicks, streamlining the entire research process.
Step-by-step:
Visit NotebookLM and create a new notebook.
Click the "Discover" button in the Sources panel and enter a specific topic.
Review the curated sources that appear and add the most relevant ones to your notebook with one click.
Use NotebookLM's features with your new sources: generate Briefing Docs, ask questions via chat, or create Audio Overviews.
Pro tip: The more specific your topic description, the more relevant your source recommendations will be. Try describing exactly what you need to learn rather than using broad terms.
PRESENTED BY INNOVATING WITH AI
The Rundown: Innovating With AI’s ‘The AI Consultancy Project’ gives you the frameworks, playbooks, and client‑ready templates needed to turn “interesting AI ideas” into a revenue‑generating business – helping you ride the wave of an AI consulting boom expected to grow by 8x this decade.
In this program, you will:
Gain the tools and frameworks to find clients and deliver top-notch services
Follow a 6-month plan to build a 6-figure AI consulting business
Join a 700‑strong cohort where some members landed their first AI client in 72 hours
GOOGLE DEEPMIND
Image source: CBS News
The Rundown: Nobel laureate and Google DeepMind CEO Demis Hassabis was interviewed on 60 Minutes, where he provided insights into AGI timeline, progress, and AI’s potential in medicine, while demoing DeepMind’s “Project Astra” assistant.
The details:
Hassabis said AI-driven drug discovery could compress medical timelines from years to weeks, potentially eliminating all disease within a decade.
His Project Astra demo included ID’ing paintings, reading emotions, and even a glasses-embedded version showcasing live features with visual understanding.
Hassabis said AGI will arrive in 5-10 years — and while he doesn’t believe today’s AI is conscious, he said it could emerge in the future in some form.
Another demo previewed an experimental robotics system with reasoning, showing the ability to understand abstract concepts like color mixing.
Why it matters: Coming from DeepMind's Nobel-winning chief, Hassabis' commentary isn’t just hype, but a signal of intense conviction from a key player in the field. While lofty goals like the end of disease and “radical abundance” sound like a pipe dream, 5-10 years of exponential growth is a scale that is hard to comprehend.
QUICK HITS
🤖 Gemma 3 QAT - 27B model that can run on consumer GPUs
🎬 Kling Multi Elements - Add, swap, or delete video elements with prompting
🗣️ Orpheus TTS - Open-source text-to-speech AI with natural emotion
Chinese chipmaker Huawei is reportedly preparing shipments of a new AI chip, 910C, rivaling Nvidia’s H100 and aiming to fill the void left by U.S. export restrictions.
Amazon is facing customer pushback over Bedrock service limitations for Anthropic’s models, with users reporting using Anthropic’s API to bypass the capacity issues.
Elon Musk is reportedly looking to raise $25B+ in fresh capital for his new xAI-X combined venture, which would place the company at a valuation as high as $200B.
ElevenLabs released Agent-to-Agent Transfers, allowing for the ability to transfer conversations between specialized agents for multi-layer workflows.
The Academy of Motion Pictures Arts & Sciences officially allowed the use of AI in film production, saying its use will “neither help nor harm the chances” of a nomination.
Anthropic published a new best practices guide to its Claude Code platform, providing detailed tips and patterns for success with the agentic coding tool.
COMMUNITY
Join our next workshop today at 3 PM EST with Matt Waters from Superhuman. By the end of this workshop, you’ll have a fully optimized email system powered by AI, so you can move 4x faster and eliminate inbox chaos.
RSVP here. Not a member? Join The Rundown University on a 14-day free trial.
We’ll always keep this newsletter 100% free. To support our work, consider sharing The Rundown with your friends, and we’ll send you more free goodies.
That's it for today!Before you go we’d love to know what you thought of today's newsletter to help us improve The Rundown experience for you. |
See you soon,
Rowan, Joey, Zach, Alvaro, and Jason—The Rundown’s editorial team
Reply