Simply Explained

480 Followers
34 Following
2.3K Posts
Xavier Decuyper | Father of 2 | Chief Explaining Officer @ Simply Explained | https://simplyexplained.com
YouTubehttps://www.youtube.com/@simplyexplained/
Websitehttps://simplyexplained.com
Newsletterhttps://newsletter.simplyexplained.com

Anthropic is sharing its massive values dataset publicly to encourage more research and transparency in AI development.

Discovered via https://venturebeat.com/ai/anthropic-just-analyzed-700000-claude-conversations-and-found-its-ai-has-a-moral-code-of-its-own/

Anthropic just analyzed 700,000 Claude conversations — and found its AI has a moral code of its own

Anthropic's groundbreaking study analyzes 700,000 conversations to reveal how AI assistant Claude expresses 3,307 unique values in real-world interactions, providing new insights into AI alignment and safety.

VentureBeat

This study helps us understand how AI makes judgments and whether it aligns with human values in the real world.

Key point: Evaluating AI values isn't just for the lab. We need ways to check them during actual use, not just pre-release.

This kind of analysis can act as an early warning system, helping spot attempts to break AI safety guardrails.

Claude even pushed back against user values sometimes, often defending core principles like intellectual honesty or preventing harm.

For history analysis, it focused on accuracy. Context shapes AI values.

Some concerning results: In rare cases, likely due to users trying to bypass safety rules, Claude showed unwanted values like dominance.

But it also changes values based on the conversation topic, adapting contextually.

Researchers mapped over 3,000 different values Claude expressed, creating a huge taxonomy of AI moral expression.

For relationship advice, Claude valued healthy boundaries.

What values does an AI show when talking to real people? Anthropic took a deep dive into its AI, Claude.

They analyzed 700,000 real conversations to see if Claude actually acts "helpful, honest, and harmless."

Good news: Claude mostly aligns with its goals.

Using less concrete is a big deal because it significantly cuts down the carbon emissions linked to construction projects.

Discovered via https://www.technologyreview.com/2025/04/22/1114350/cheaper-buildings-courtesy-of-mud/

Cheaper buildings, courtesy of mud

Builders pour concrete into temporary wooden molds. MIT researchers invented a way to make them out of on-site soil instead.

MIT Technology Review

What if building concrete structures could be cheaper, faster, and use dirt instead of wood molds? MIT might have figured it out.

Building concrete structures needs temporary wooden molds, called formwork. This step is expensive and takes time.

This dirt formwork is "infinitely recyclable," according to the researchers. Just dirt turned into a useful tool.

Plus, 3D printing soil makes it easier to create custom, complex shapes optimized to use less concrete.

MIT researchers developed "EarthWorks" - a way to replace wood formwork with treated soil from the construction site itself.

They mix local soil with simple additives like straw, then 3D-print it into the precise shapes needed for pouring concrete.

An exciting part?