Praful Hambarde

LLMs forget critical context in multi-turn conversations not because they lack aptitude, but because they can't distinguish important turns from irrelevant ones – and a simple relevance-weighted history fixes it.

A unified benchmark reveals the trade-offs between pixel-wise accuracy and perceptual realism in state-of-the-art image super-resolution techniques.

Even state-of-the-art AI-generated image detectors struggle when images are cropped, resized, or compressed, revealing a critical gap in real-world robustness.