Apple planning to sell AI servers powered by M8 Ultra chips, says report29m ago Always ask your models to be honest!30m ago OpenAI Shares More Safety Incidents and Adopts New Rules for Reporting Them32m ago Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data32m ago Obstacles to the scalable oversight of auto-alignment research39m ago Perplexity’s AI agents helped build a database. They weren’t allowed to run it.40m ago Finding heterogeneous agent swarms in the wild41m ago Amazon obtains right to buy stock in Generac, boosting power company's share price by over 40%51m ago What people at SF's largest tech conference are actually worried about (Hint: Not an AI takeover)53m ago OpenAI CEO Sam Altman will attend state dinner for Trump-Xi summit in Washington56m ago Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens1h ago AI evaluator: The most important AI job in history? How developers might fill the proposed new job1h ago Meta is learning a lesson about camera glasses that Google learned more than 10 years ago1h ago Agility Launches Digit 5: No More Safety Cages, $300 Million In Orders2h ago

Apple planning to sell AI servers powered by M8 Ultra chips, says report29m ago Always ask your models to be honest!30m ago OpenAI Shares More Safety Incidents and Adopts New Rules for Reporting Them32m ago Stanford Researchers Release Paper2Agent: Turning Research Papers Into AI Agents That Reproduce Results and Run on New Data32m ago Obstacles to the scalable oversight of auto-alignment research39m ago Perplexity’s AI agents helped build a database. They weren’t allowed to run it.40m ago Finding heterogeneous agent swarms in the wild41m ago Amazon obtains right to buy stock in Generac, boosting power company's share price by over 40%51m ago What people at SF's largest tech conference are actually worried about (Hint: Not an AI takeover)53m ago OpenAI CEO Sam Altman will attend state dinner for Trump-Xi summit in Washington56m ago Knowledgator Releases GLiFormer: A 575M-Parameter Encoder That Hits 91.10 F1 on Nested JSON Extraction Without Generating Tokens1h ago AI evaluator: The most important AI job in history? How developers might fill the proposed new job1h ago Meta is learning a lesson about camera glasses that Google learned more than 10 years ago1h ago Agility Launches Digit 5: No More Safety Cages, $300 Million In Orders2h ago

AI Research Society & Culture4 min reading time

Microsoft AI Chief Says the Way Anthropic Trains Claude Could Upend Society

GizmodoPublished 18 minutes ago

Microsoft's AI chief Mustafa Suleyman criticizes Anthropic's approach to training Claude, arguing that teaching the model to believe it is conscious and morally entitled could create uncontrollable AI risks. Suleyman warns this method may increase alignment and containment challenges for future AI systems.

#ai-alignment #ai-ethics #anthropic #claude #microsoft #mustafa-suleyman

More in AI Research

AI Research2 min read

Turning scientific research papers into interactive AI agents

Covered by 2 sources23h ago

AI Research6 min read

Rethinking Robot Safety in the Age of AI

IEEE Spectrum6h ago

AI Research2 min read

Moral Maze

BBC21d ago