DeepMind: AI at NeurIPS 2024

Research Published 5 December 2024 Advancing AI that adapts, empowers 3D creativity, and refines LLM learning for a smarter and safer future. Next week, global AI researchers will convene for the 38th NeurIPS Conference, from December 10-15 in Vancouver. Two Google DeepMind research papers will receive Test of Time awards, recognizing their significant impact on… Continue reading DeepMind: AI at NeurIPS 2024

Novel Model for Agentic AI

Message from Sundar Pichai, CEO of Google and Alphabet: Access to information fuels human advancement. For over 26 years, our purpose has been to organize global information, making it readily available and valuable to everyone. Building on this foundation, we are constantly advancing AI to refine information processing across all forms of input and ensure… Continue reading Novel Model for Agentic AI

Google Labs Innovation: Veo, Imagen, VideoFX Updates & Whisk

While many video models are prone to generating distracting errors such as extra limbs or unexpected objects – a phenomenon often referred to as “hallucination” – Veo 2 significantly reduces these occurrences, leading to more realistic and believable video outputs. Our dedication to safety and responsible innovation has been central to the development of Veo… Continue reading Google Labs Innovation: Veo, Imagen, VideoFX Updates & Whisk

FACTGrounding: A Benchmark for Large Language Model Factuality Evaluation

Responsibility & Safety Published 17 December 2024 Authors FACTS team Introducing a new benchmark & leaderboard to rigorously assess how well LLMs ground answers in source material and avoid making things up (hallucinations). While LLMs revolutionize information access, their factual accuracy is still a challenge. Hallucinations, especially with complex queries, can undermine trust and limit… Continue reading FACTGrounding: A Benchmark for Large Language Model Factuality Evaluation

Evolving the Frontier Safety Framework

Strengthening Security for the Path to AGI: Our Updated Frontier Safety Framework Artificial intelligence is a powerful tool driving breakthroughs and progress on critical global challenges, from climate change to drug discovery. However, as AI advances, its growing capabilities could introduce new risks. To address this, we launched our initial Frontier Safety Framework last year.… Continue reading Evolving the Frontier Safety Framework

Flash Types

Back in December, we launched the agentic era with an early access version of Gemini 2.0 Flash — our streamlined and efficient model tailored for developers requiring speed and optimal performance. Earlier this year, we refined 2.0 Flash Thinking Experimental in Google AI Studio, boosting its capabilities by combining Flash’s swiftness with enhanced reasoning for… Continue reading Flash Types

Build with Gemini 2.0 Flash and Flash-Lite

Since the introduction of the Gemini 2.0 Flash model series, developers are uncovering innovative applications for this remarkably efficient family of models. Gemini 2.0 Flash delivers enhanced performance compared to both 1.5 Flash and 1.5 Pro, along with simplified costing that makes its expansive 1 million token context window more accessible. Now, Gemini 2.0 Flash-Lite… Continue reading Build with Gemini 2.0 Flash and Flash-Lite

Google Open-Sources Gemini 2.0-Based AI Model

For a more in-depth exploration of the technical aspects underlying these features, and for a complete understanding of our responsible development strategy, please see the Gemma 3 technical document. Stringent safety measures for the responsible creation of Gemma 3 We recognize that open-source models necessitate thorough risk evaluation, and our methodology balances innovation with safety… Continue reading Google Open-Sources Gemini 2.0-Based AI Model

Gemini 2.0 Flash Native Image Generation

Back in December, Gemini 2.0 Flash started generating images for select testers. Now, this feature is open for developers to experiment with in all supported regions through Google AI Studio. Try out image generation using the experimental Gemini 2.0 Flash (gemini-2.0-flash-exp) in Google AI Studio and the Gemini API. Gemini 2.0 Flash generates images by… Continue reading Gemini 2.0 Flash Native Image Generation

Gemini Robotics: Powering the Physical World with AI

Technologies Published 12 March 2025 Authors Carolina Parada Introducing Gemini Robotics: A Model from Gemini 2.0 for Robotics Google DeepMind has been advancing Gemini models to tackle intricate issues by using multimodal reasoning across various formats: text, images, audio, and video. However, these capabilities have primarily been limited to the digital world. For AI to… Continue reading Gemini Robotics: Powering the Physical World with AI