Tag: #multimodal-llm14 results
Apple Machine Learning examines how preference alignment affects multimodal large language models (MLLMs), an area that remains less explored than language-model alignment. The study focuses on reducing hallucinations, including responses that conflict with image content.
Ghelia has launched a new platform that uses multimodal LLMs to integrate 3D models and analysis results. It aims to transform manufacturing failure data and tacit design knowledge into structured, searchable organizational assets.
Speech-adapted LLMs underperform text-based counterparts and cascaded pipelines on language tasks, termed the text-speech understanding gap. This performance drop occurs when processing spoken inputs versus equivalent text. Recent gap-narrowing methods rely on costly large-scale speech synthesis.
Dongfeng Auto's self-developed Dongfeng Taiji large model passed national generative AI service filing. Enterprise-grade multimodal model enables full-scenario synergy across auto product dev, ops, manufacturing, sales. Built on Qiantian compute platform and proprietary datasets.