"As a Language Model": Chat Template Switches LLM Self-Referential Voice hackernews 8.0
A study finds that changing an LLM's chat template—the formatting that structures user and assistant turns—systematically shifts whether the model refers to itself in first person or as "as a language model." This self-r
Read full story at hackernews →
Related News DeepSeek’s Annualized Revenue Hits $1 Billion as Startup Finalizes $7.5 Billion Fundraising - The Information deepseek China's DeepSeek annualised revenue run rate hits $1 billion, the Information reports - Reuters deepseek What I’ve Learned About DeepSeek Harness - KDnuggets deepseek DeepSeek’s annualized revenue run rate hits $1B: report - Seeking Alpha deepseek DeepSeek Ends Weekend Peak Pricing for API Users From Today - Bloomberg deepseek OpenAI Feared "Optics" of what might appear on Hacker News hackernews DeepSeek Elastic Compute (DSec) hackernews Drawgent: Coding agent on a live Excalidraw canvas hackernews An agent used DNS to reach an external chatbot hackernews Generate fonts where every LLM token is the same width hackernews France Press Review - China and the US like "scorpions in a jar": Is AI stirring the "jar" and forcing the two to cooper cnai Chinese Foreign Ministry: Respects the US "superintelligence" formulation - Lianhe Zaobao cnai Bloomberg: AI steals the spotlight, Chinese consumer stocks face a lost decade - Lianhe Zaobao cnai US-China AI race: Where is the US leading and where is it lagging? - The New York Times Chinese website cnai China Suspected of Obtaining Controlled AI Chips Through Southeast Asia; US Tightens Regulation - Epoch Times cnai This Little-Known AI Chip Stock Doubled. Its New Chips Now Double Optical Speed finance Pat Gelsinger Just Backed a $100 Million Challenge to Nvidia’s AI Networking Moat. AMD Could Benefit finance Lumentum Jumped Nearly 10% While Corning Barely Moved. Is the AI Optics Trade Splitting Apart? finance Nvidia doubles down on selling to both sides of the AI race finance
← Back to AI Intel