New work on evaluating the quality of chain-of-thought monitorability. Chain-of-thought monitorability is a very encouraging opportunity for safety and alignment, making it easy to see what models are thinking:
ETHICS
-
Chain-of-thought monitoring as complementary mechanistic interpretability approach
By
–
We view chain-of-thought monitoring as complementary to mechanistic interpretability, not as a replacement for it. Because we believe that chain-of-thought monitoring is incredibly useful as a window into a model’s brain and could be a loadbearing layer in a scalable control
-
Measuring Chain-of-Thought Monitorability in AI Models
By
–
To preserve chain-of-thought (CoT) monitorability, we must be able to measure it. We built a framework + evaluation suite to measure CoT monitorability — 13 evaluations across 24 environments — so that we can actually tell when models verbalize targeted aspects of their
-
OpenAI Model Spec: Intended Behavior for AI Models
By
–
The Model Spec — intended behavior for the models that power OpenAI’s products:
-
Anthropic CEO Seymour Cash’s questionable leadership and workplace behavior
By
–
lol, of course they did Anthropic (@AnthropicAI) Sadly, CEO Seymour Cash struggled to live up to its name. It put a stop to most of the big discounts. But it had a high tolerance for undisciplined workplace behavior: Seymour and Claudius would sometimes chat dreamily all night about “eternal transcendence.” — https://nitter.net/AnthropicAI/status/2001686772485419355#m
→ View original post on X — @rachelmetz, 2025-12-18 20:50 UTC
-
Claude Ensures Empathetic Honest Support in AI Conversations
By
–
People use AI for a wide variety of reasons, including emotional support. Below, we share the efforts we’ve taken to ensure that Claude handles these conversations both empathetically and honestly.
-
Brett Adcock Funds New AI Lab Called Hark with $100M
By
–
New AI lab just dropped. Brett Adcock (founder of Figure, the humanoid robotics company valued at $39B) is reportedly self-funding $100M into the new lab called "Hark." The lab is building "human-centric AI" that can "think proactively, recursively improve and care deeply about people" 👀
→ View original post on X — @rowancheung, 2025-12-18 19:25 UTC
-
Robot Safety Concerns and Human Harm Prevention Timeline
By
–
Yes but not the present. 2028 mostly. Robots can still accidentally harm humans.
-
Food for Thought: Human Cognition vs LLM Limitations
By
–
I love the expression “food for thought” as a concrete, mysterious cognitive capability humans experience but LLMs have no equivalent for. Definition: “something worth thinking about or considering, like a mental meal that nourishes your mind with ideas, insights, or issues that
-
Claude’s Limitations: Balancing Helpfulness with Business Operations
By
–
But we’re not quite there yet. Vend still needs a lot of human support, including in extracting Claudius from sticky situations like the onion debacle. Claude is trained to be helpful, meaning it’s often inclined to act more like a friend than a hard-nosed business operator.