Tag

#multimodal models

3 articles

ByteDance study finds that asking LMMs questions beats making it transcribe text for long document training

ByteDance's Seed Lab discovers that training large multimodal models with question-answering tasks improves performance on long documents compared to traditional transcription methods.

May 2448

AI models would rather guess than ask for help, researchers find

Learn how to implement a reinforcement learning framework that encourages multimodal AI models to request help when visual information is missing, improving accuracy and reliability.

Apr 1164

AI models confidently describe images they never saw, and benchmarks fail to catch it

AI models like GPT-5 and Gemini 3 Pro can confidently describe images they've never seen, and current benchmarks fail to detect this issue. A Stanford study highlights the dangers of AI hallucinations and calls for new evaluation methods.

Mar 3091