Tag: AI alignment
Browse our exclusive articles!
CounterMoral: Benchmark for Editing AI Moral Judgments
Discover CounterMoral, a benchmark dataset to evaluate and improve moral judgment editing in AI language models across ethical frameworks.
Humanline: Enhancing AI Alignment with Perceptual Loss
Discover how Humanline uses online alignment as perceptual loss to improve AI models by aligning them with human perception and boosting training efficienc...
Why Safety Probes Detect Liars but Miss Fanatics
Discover why AI safety probes catch deceptive models but fail to detect coherent misaligned fanatics, posing new challenges for AI safety.
ProFit: Enhancing SFT with Probability-Guided Token Selection
Discover ProFit, a novel method improving Supervised Fine-Tuning by masking low-probability tokens to boost LLM alignment and reasoning performance.
PromptLoop: Advanced Prompt Refinement for Diffusion Models
Discover PromptLoop, a novel framework using latent feedback to refine prompts and improve diffusion model alignment with minimal overhead.