Alham Fikri Aji

Most web-browsing LLM agents struggle to find obscure, multi-hop information across diverse modalities and languages, even when explicitly designed for web interaction.

Current egocentric video generation models fail to execute multi-step manipulation tasks realistically, often skipping crucial actions and exhibiting poor physical dynamics.

Machine translation benchmarks have functionally saturated, but pairing human-authored failure cases with deterministic verification rules reveals critical multimodal blind spots that automated metrics consistently miss.

EDRAC reveals that existing Arabic LLMs struggle with dialectal fidelity, exposing a critical gap in current benchmarks for Arabic NLP.

Now you can retrofit any frozen language model with explicit, editable "sense knobs" for steering, disambiguation, and cross-lingual transfer, without pretraining from scratch.

Fixing your parallelism strategy while tuning batch size (or vice versa) leaves performance on the table: COPUS adaptively co-tunes both for faster LLM training.

LLMs still struggle to reason in context when cultural and linguistic nuances are involved, achieving only 44% accuracy on a new grounded benchmark spanning 14 languages.

Fine-tuning VLMs for regional relevance doesn't have to sacrifice global performance: a simple data filtering and model merging technique boosts cultural relevance by 5-15% while barely impacting overall accuracy.

VLMs can regain lost linguistic prowess without extra parameters or architectural changes, thanks to a clever KV-cache sharing trick for distillation.