New Research Links AI Sycophancy to Feature Geometry in LLMs
New research explains AI sycophancy and misalignment through feature superposition geometry, while OpenAI deploys specialized cybersecurity…
Breaking news, research, and analysis on artificial intelligence — covering the models, methods, and shifts shaping the field.
New research explains AI sycophancy and misalignment through feature superposition geometry, while OpenAI deploys specialized cybersecurity…
Poolside launched Laguna XS.2, a free open-source AI model optimized for coding workflows, challenging expensive proprietary…
Sakana AI published research on its RL Conductor, a 7B-parameter model that automatically orchestrates GPT-5, Claude…
Sakana AI released RL Conductor, a 7B parameter model that orchestrates frontier LLMs like GPT-5 and…
TokenArena introduces energy-aware AI benchmarking across 78 endpoints, revealing 6.2x efficiency variations and 12.5-point accuracy differences…
Zyphra released ZAYA1-8B, an 8.1 billion parameter model with only 760 million active parameters that matches…
AI reasoning capabilities advanced significantly this week with xAI's Grok 4.3 launching at aggressive $1.25 per…
Sakana AI's RL Conductor uses a 7B parameter model to automatically orchestrate GPT-5, Claude Sonnet 4,…
New research reveals AI reasoning models consume dramatically more compute while hitting performance ceilings on creative…
xAI launched Grok 4.3 at $1.25 per million input tokens, significantly undercutting competitors while showing domain-specific…
Krutrim, India's first AI unicorn, pivots from model development to cloud services after workforce cuts and…
Sakana AI released RL Conductor, a 7B parameter model that automatically orchestrates GPT-5, Claude, and other…