GPT‑4V, Gemini praised
Multimodal foundation models like GPT‑4V and Gemini are being singled out for cross‑domain reasoning — people are citing applications in healthcare diagnostics and autonomous systems as immediate use cases. (x.com)
A large-scale evaluation on arXiv tested GPT‑4V across 16 medical‑imaging categories — including radiology, oncology, ophthalmology and pathology — and reported strengths in modality recognition, anatomy localization, disease diagnosis and lesion detection. (arxiv.org) A separate peer‑reviewed effort assessed GPT‑4V’s autonomous diagnostic performance on 206 imaging studies made up of 60 radiographs, 60 CT scans, 60 MRIs and 26 angiographies. (link.springer.com) Google Research published a Med‑Gemini announcement describing a family of multimodal clinical models built on Gemini that use self‑training, web‑search integration and fine‑tuning to improve clinical reasoning and multimodal performance. (research.google) “On the Road with GPT‑4V,” an ICLR/arXiv evaluation, probed GPT‑4V as an autonomous‑driving agent and found the model notably improved scene understanding and causal reasoning for driving scenarios. (arxiv.org) A quantitative autonomous‑driving analysis reported GPT‑4V achieved about 57% zero‑shot accuracy predicting pedestrian crossing actions, compared with roughly 70% for state‑of‑the‑art domain‑specific models. (proceedings.aaai-make.info) Waymo published research introducing EMMA, an end‑to‑end multimodal model that leverages Gemini’s capabilities to generate driving trajectories, and the company released the EMMA paper in October 2024. (waymo.com) Automakers and suppliers have moved from experiments to pilots: Volvo demonstrated Gemini integration in the EX90 on June 25, 2025, and General Motors has announced plans to integrate Gemini‑powered features with ambitions for “eyes‑off, hands‑off” driving by 2028. (applyingai.com)
