Bahador Rashidi

Multi-Head Latent Control reduces large model usage by up to 90.7% while enhancing decision-making accuracy in LLM agents.

LLMs struggle with raw visual tool outputs, but a simple trick—translating pixels into language-native summaries—unlocks massive gains in visual reasoning, outperforming even fine-tuned agents.

Current MLLMs can't see the forest for the trees when it comes to image quality assessment, failing to understand region-level degradations even with explicit cues.