Three Years from GPT-3 to Gemini 3

Alphabet Inc. shares increased by as much as 6.9% on Wednesday, marking the largest gain since early September. This surge in stock price follows a series of positive reviews for the company's latest AI model, Gemini 3, which has been praised for its advancements in coding and reasoning capabilities.

Oriol Vinyals, co-lead of Gemini, stated that the advancements in Gemini 3 are attributed to improved pre-training and post-training processes. This statement challenges the prevailing notion that gains from pre-training are diminishing. Vinyals emphasized the importance of these enhancements in the context of AI development.

Google's latest AI model, Gemini 3, reportedly surpasses competitors in various benchmark tests. However, concerns about its reliability persist, raising questions about the sustainability of the current AI hype.

Google has launched Gemini 3, its most advanced AI model to date, which promises enhanced reasoning capabilities, 'vibe coding', and the ability to perform complex multi-step tasks. This development raises questions about the potential impact of AI on human jobs and the future of work.

Demis Hassabis, CEO of Google DeepMind, discussed the advancements of Gemini 3, Google's latest AI model, in an exclusive interview. The model is positioned as a significant leap in AI technology, promising enhanced reasoning and coding capabilities. This development marks a pivotal moment as Google enters a new era of intelligence with Gemini 3.

The paper introduces AdamHuberDecay, a new adaptive optimizer designed for pre-training large transformer-based generative models. This optimizer replaces the traditional $ ext{l}_2$ penalty used in AdamW with a decoupled smooth Huber regularizer. The proposed method allows for quadratic decay of parameters below a certain threshold while applying linear decay for larger values. This approach aims to improve regularization gradients, maintain invariance to second-moment rescaling, and enhance sparsity for overgrown weights.

Demis Hassabis, co-founder of DeepMind, discussed the advancements of Gemini 3, Google's latest AI model, emphasizing its capabilities in world modeling and fitting the entire Google Search index into the system. He addressed concerns about the AI bubble and highlighted the model's potential to enhance user interactions and provide more accurate information. Hassabis's insights reflect a commitment to pushing the boundaries of AI technology and its integration into everyday applications.

Gemini 3, Google's latest AI model, showcases significant advancements in planning, coding, and judgment skills. The model has evolved from previous iterations, moving beyond hallucinations to exhibit subtle, human-like errors. This development indicates a notable improvement in AI capabilities, enhancing user interactions and providing more accurate responses.