Core idea: We can treat rubrics like models. Optimize along two axes: (1) Alignment with stakeholder objectives (2) Agreement among evaluators, both human and LLMAJ
SAFETY
-
Majority Not Always Right: RL Training for Solution Aggregation
By
–
The Majority is not always right RL training for solution aggregation
-
Western AI Censoring Causes More Production Issues Than Chinese Models
By
–
For a production app the Western AI models censoring is much more annoying than the Chinese censoring In 0.0001% of cases someone will try to generate Tiananmen Square But in 80% of cases they generate a normal video which gets false flagged as NSFW with Veo 3 for example https://
t.co/RfK54OMvSC -
Veo3 Video Model Plagued by Excessive NSFW False Positives
By
–
Veo3's video model is still completely unusable You can't generate basic things with it because 80% gets false flagged as NSFW
-
Hinton warns superintelligent AI will cause massive job losses
By
–
The godfather of AI, Geoffrey Hinton: „But when you get superintelligent AI, its not clear what jobs those people are going to do. So I believe that we are going to see fairly a massive loss of jobs.“ https://t.co/zNxtgtsrVX
— Chubby♨️ (@kimmonismus) 11 septembre 2025The godfather of AI, Geoffrey Hinton: „But when you get superintelligent AI, its not clear what jobs those people are going to do. So I believe that we are going to see fairly a massive loss of jobs.“
-
Stuart Russell Named TIME100AI for AI Ethics Leadership
By
–
BAIR Faculty and co-founder of the International Association for Safe and Ethical AI, Stuart Russell, is recognized as one of this year's #TIME100AI.
-
Practical Rationality: Beyond Suboptimal AI Integration
By
–
it is irrational to submit to a tool with suboptimal outcomes; the proper integration of perception, intuition, feeling and reasoning is part of practical rationality
-
Fighting AI Doomerism A Hunger Strike Against Pessimism
By
–
day 1 of my hunger strike against the hunger strike of the doomers
-
Security guardrails for coding agents in organizations
By
–
We need more visibility into their actions and be able to set real-time guardrails across all our coding agents to ensure we're security compliant. If you're rolling out Claude Code to your organization, this is crucial.