Google says its Gemini AI outperforms both GPT-4 and expert humans - 3 Quarks Daily
Google’s new Gemini models demonstrate significant leaps in AI capability, with the Ultra version reportedly surpassing human performance on complex, multi-domain reasoning benchmarks. This achievement highlights the rapidly evolving frontier of artificial intelligence, moving beyond simple pattern recognition to genuine expert-level understanding. The existence of varied versions, including lightweight models for mobile devices, suggests a strategic push toward accessibility alongside high-end performance. By offering scalable options, Google aims to democratize advanced AI tools while maintaining a competitive edge against rivals like OpenAI and Meta. This development is crucial for open data communities as it raises questions about transparency and reproducibility in state-of-the-art AI. With Google withholding details on training data and model sizes, it becomes harder for researchers to audit these systems. Consequently, this lack of openness hinders independent verification and limits the broader scientific community’s ability to build upon these advancements responsibly.
Source: 3quarksdaily.comPublished on 2023-12-08