Hisham Cholakkal

A single visual world model can now execute complex procedural tasks end-to-end, adapting its generation to the actual outcome of each predicted action.

A 0.9B parameter omni-modal embedder achieves state-of-the-art text retrieval quality by distilling a frozen text backbone's embeddings into new modalities, bypassing the common modality-extension performance drop.

Frontier generalist systems have effectively mastered high-level visual semantics and spatial reasoning, yet they still stubbornly fail to displace dedicated models on metric geometry, dense temporal tracking, and fine-grained visual fidelity.

Visual under-conditioning in LMMs can be overcome by directly regularizing visual attention, leading to remarkable improvements in multimodal understanding tasks.

Unsupervised training can elevate multimodal models, achieving a 3.5% boost in understanding metrics and a notable increase in image generation fidelity without human intervention.

Navigating the deluge of scientific papers just got easier: Paper Circle uses multi-agent LLMs to automate research discovery and knowledge graph construction, promising faster literature reviews and deeper insights.

Forget choosing just one vision encoder – fusing CLIP and DINO representations unlocks a significant performance boost in vision-language tasks.

Open-ended reinforcement learning with LLM-based rewards unlocks surprisingly strong performance in medical reasoning for multimodal models, even with limited training data.

Forget cloud GPUs – a new model brings unified multimodal understanding and generation to your iPhone, running 6x faster than alternatives.