Committed to Responsible AI Development

We at Google firmly believe that AI has the power to transform lives and contribute to a better world, provided it's developed with boldness and responsibility. Our AI technologies are already making a difference, helping people with everyday tasks and supporting them in their most ambitious, creative, and productive endeavors.

To us, responsible AI development means tackling potential risks while also maximizing the benefits for individuals and society at large. We are guided by our AI Principles, our ongoing research, and valuable input from experts, users, and partners.

Building AI to Benefit Society with LearnLM

Every day, billions of people turn to Google to learn. With the advent of generative AI, we're unlocking new, personalized, helpful, and accessible ways to enhance learning.

We're excited to introduce LearnLM, a new family of models built on the foundation of Gemini, and specifically fine-tuned for educational purposes. LearnLM incorporates research-backed learning science and academic principles into our products, helping to manage cognitive load and adapt to each learner's unique goals, needs, and motivations. This results in a learning experience that's tailored to the individual.

[ttpp]LearnLM is powering a range of features across our products, including Gemini, Search, YouTube, and Google Classroom. In the Gemini app, the new Learning coach Gem provides step-by-step study guidance, designed to foster understanding rather than just delivering answers. On the YouTube app for Android, users can now interact with educational content in new ways, such as asking follow-up questions or taking quizzes. Thanks to the long-context capability of the Gemini model, these features work seamlessly even with long lectures and seminars.[yyxx]

Collaboration is key to realizing the full potential of generative AI in education. We're partnering with leading educational institutions like Columbia Teachers College, Arizona State University, NYU Tisch, and Khan Academy to refine our models. Our aim is to extend the reach of LearnLM beyond our own products. If you're interested in partnering with us, you can sign up here. We've also joined forces with MIT RAISE to create an online course that equips educators with tools to integrate generative AI effectively in their teaching.

Using Long-Context to Make Knowledge Accessible

We've developed an experimental tool called Illuminate, which uses the long-context capabilities of Gemini 1.5 Pro to convert complex research papers into concise audio dialogues. In just minutes, Illuminate can generate a conversation between two AI-generated voices, offering a summary and key insights from research papers. For those who want to delve deeper, follow-up questions are supported. All audio is watermarked with SynthID, and the original papers are referenced for further exploration. You can sign up to try Illuminate today at labs.google.

[ttpp]While these innovations help us fulfill our mission in new ways, generative AI is still evolving, and with it come risks and questions that need addressing.

We're committed to a responsible approach to AI, guided by our AI Principles. We regularly share updates on our learnings and practices.

Today, we're highlighting a few new ways we're enhancing our models like Gemini and safeguarding against misuse:

• AI-Assisted Red Teaming and Expert Feedback: We're enhancing our red teaming practices, which involve proactively testing our systems for vulnerabilities, with a new technique called "AI-Assisted Red Teaming." Inspired by Google DeepMind's gaming breakthroughs like AlphaGo, we train AI agents to compete and expand their red teaming capabilities. This helps address adversarial prompting and limit problematic outputs. We also incorporate feedback from thousands of internal safety specialists and independent experts from academia to civil society, enhancing our models' accuracy and reliability.

• SynthID for Text and Video: As our model outputs become more realistic, we're expanding SynthID to include text and video, in addition to images and audio. This technology adds imperceptible watermarks to help identify AI-generated content and prevent misuse. We're committed to helping people understand the origin of digital content.

• Collaborating on Safeguards: We're working with the broader ecosystem to share and improve our advances. In the coming months, we'll open-source SynthID text watermarking through our updated Responsible Generative AI Toolkit. We're also part of the Coalition for Content Provenance and Authenticity (C2PA), collaborating with Adobe, Microsoft, startups, and others to enhance transparency in digital media.[yyxx]

Helping to Solve Real World Problems

Our AI advances are already helping tackle real-world challenges and drive scientific breakthroughs. Recently, we've announced several exciting milestones:

• In collaboration with Harvard University, Google Research published the largest synaptic-resolution 3D reconstruction of the human cortex, which could revolutionize our understanding of brain function.

• We introduced AlphaFold 3, an update to our model that now predicts the structure and interactions of DNA, RNA, and ligands, in addition to proteins. This advancement could transform our understanding of biology and drug discovery.

• We unveiled Med-Gemini, a family of research models that leverages Gemini's capabilities in advanced reasoning, multimodal understanding, and long-context processing. These models could assist clinicians with administrative tasks, analyze various medical data, and aid in risk prediction.

These achievements add to the societal benefits we've seen from projects like flood forecasting and the UN Data Commons for the SDGs. We recently published a report detailing how AI can help advance our progress on the world's shared Sustainable Development Goals. We're just getting started, and we're excited about the future and the potential impact of our collaborative efforts.

[ttpp]Get more stories from Google in your inbox.

Email addressYour information will be used in accordance with Google's privacy policy.

SubscribeDone. Just one step more.

Check your inbox to confirm your subscription.

You are already subscribed to our newsletter.

You can also subscribe with a different email address.[yyxx]