alt Hacker NewsClaude Sonnet 4.6
Comments
Your best guess for the Sonnet family number of parameters? 400b?
On a passive aggressively prompted AI:
> I want to wash my car. The car wash is 50 meters away. Should I walk or drive?
Walk. It will give you time to think about why you need an AI to answer such obvious questions.
great stuff
The scary implication here is that deception is effectively a higher order capability not a bug. For a model to successfully "play dead" during safety training and only activate later, it requires a form of situational awareness. It has to distinguish between I am being tested/trained and I am in deployment.
It feels like we're hitting a point where alignment becomes adversarial against intelligence itself. The smarter the model gets, the better it becomes at Goodharting the loss function. We aren't teaching these models morality we're just teaching them how to pass a polygraph.
[dead]
[dead]
[flagged]
Whoa, I think Claude Sonnet 4.5 was a disappointment, but Claude Sonnet 4.6 is definitely the future!
The best, and now promoted by the US government as the most freedom loving!
Look at these pelicans fly! Come on, pelican!
It excels at agentic knowledge work. These custom, domain-specific playbooks are tailor made: claudecodehq.com
https://www.anthropic.com/news/claude-sonnet-4-6
The much more palatable blog post.