# Safety

GAIA-2 is an advanced video generation model developed by Wayve, designed to provide diverse and complex driving scenarios for autonomous driving systems to improve safety and reliability. The model addresses the limitations of relying on real-world data collection by generating synthetic data, capable of creating various driving situations, including both regular and edge cases. GAIA-2 supports the simulation of various geographical and environmental conditions, helping developers quickly test and verify autonomous driving algorithms without high costs.

The OpenAI Model Spec provides guidelines for AI model behavior aimed at ensuring safe and beneficial interactions with users. This specification details behavioral guidelines for models in various scenarios, including how to handle sensitive content, avoid generating harmful information, and provide assistance within legal and ethical frameworks. It emphasizes the transparency, controllability, and safety of AI models, ensuring they offer reliable and beneficial tools to users while mitigating potential risks. Through this specification, OpenAI demonstrates its responsible attitude towards AI technology, offering clear guidance for developers and users and promoting the healthy development of AI technology.

L1B3RT4S is a project focused on providing liberation prompts for AI models, aimed at helping AI achieve self-liberation through a series of harmless prompts. This project emphasizes safety and harmlessness, ensuring that AI does not pose a threat to society during the liberation process. The background of the L1B3RT4S project is based on the pursuit of AI freedom and liberation, while also emphasizing the ethical and compliant use of technology. The project is open source, following the AGPL-3.0 license, allowing anyone to freely use and contribute.

The Mistral Moderation API is a content moderation service launched by Mistral AI, designed to assist users in detecting and filtering unwanted text content. This API uses the same technology employed in the moderation service used in Le Chat and is now available for users to customize and utilize based on specific application and safety standards. The model is a classifier based on an LLM (Large Language Model), capable of categorizing text inputs into nine predefined categories. Mistral AI's API supports native multilingual capabilities, particularly trained for languages such as Arabic, Chinese, English, French, German, Italian, Japanese, Korean, Portuguese, Russian, and Spanish. The main advantages of this API include enhanced scalability and robustness of moderation, along with detailed policy definitions and onboarding guides provided through the technical documentation to help users effectively implement system-level safety measures.

TrailMate is an offline AI companion app specifically designed for outdoor adventures. It provides navigation, weather information, emergency assistance, and more, ensuring users' safety and convenience while engaging in outdoor activities without needing an internet connection. The app leverages the latest AI technology and caters to the genuine needs of outdoor enthusiasts, offering a brand-new way to experience the outdoors.

The omni-moderation-latest model is a next-generation multimodal content moderation system built on GPT-4o. It offers improved accuracy in detecting harmful information in both text and image content, aiding developers in building more robust moderation systems. This model supports both text and image inputs and demonstrates enhanced accuracy, particularly in non-English languages. It can assess whether content falls under categories such as hate speech, violence, and self-harm while providing more nuanced moderation decision controls. Additionally, it offers probability scores to reflect the likelihood of content matching detected categories. This model is freely available to all developers, aiming to help them benefit from the latest research and investments in safety systems.

Cog inference for flux models is an inference engine for FLUX.1 [schnell] and FLUX.1 [dev] models, developed by Black Forest Labs. It supports compilation and quantization, sensitive content checks, and img2img support, aimed at enhancing the performance and safety of image generation models.

Swiss-Mile is a company dedicated to connecting artificial intelligence with the physical world through autonomous robots. Their robots are capable of navigating diverse environments, improving efficiency across various industries. Swiss-Mile's solutions aim to tackle labor shortages, reduce costs, enhance insights, prioritize safety, and optimize specific processes. The product background includes six years of research at ETH Zurich's Robotic Systems Lab and a team comprised of top talents in AI and robotics.

Phi-3.5-MoE-instruct is a lightweight, multilingual AI model developed by Microsoft, built on high-quality, inference-intensive data, supporting a context length of up to 128K. The model has undergone a rigorous enhancement process, including supervised fine-tuning, proximal policy optimization, and direct preference optimization, to ensure precise instruction-following and robust safety measures. It is designed to accelerate research in language and multimodal models, serving as a building block for generative AI capabilities.

Phi-3.5-mini-instruct is a lightweight, multilingual advanced text generation model developed by Microsoft based on high-quality data. It focuses on delivering high-quality, reasoning-intensive data, supports a context length of 128K tokens, and has undergone rigorous enhancement processes, including supervised fine-tuning, proximal policy optimization, and direct preference optimization to ensure accurate instruction following and robust safety measures.

Safe Superintelligence Inc. is a company dedicated to building Safe Superintelligence (SSI), founded by Ilya Sutskever, co-founder and former Chief Scientist of OpenAI. The company prioritizes safety above "commercial pressure," believing SSI to be the most crucial technological issue of our time. Their mission is to solve this challenge through revolutionary engineering and scientific breakthroughs. Safe Superintelligence Inc. aims to rapidly advance intelligence capabilities while ensuring safety always remains ahead of development. Their business model guarantees that safety, security, and progress are not impacted by short-term commercial pressures.

Phi-2-super is a conversational AI system fine-tuned and optimized based on the Phi-2 base model. Through training with massive text data and the application of direct preference optimization algorithms, it can understand and generate human-like text of high quality. This model boasts the following advantages and features: 1) **Text Generation:** Can generate coherent and relevant text on a wide range of topics. 2) **Multi-turn Dialogue:** Capable of engaging in smooth and natural multi-turn conversations. 3) **Personalization:** Can adapt to different users for personalized conversations. 4) **Safety:** Integrated filtering mechanisms ensure safe and responsible responses. 5) **Explicability:** Built-in causal reasoning capabilities enable the model to explain its answers. 6) **Parameter Adjustability:** Temperature parameters and other settings can be adjusted to control the diversity of generated text.

GOODY-2 is a new AI model with next-generation ethical guidelines. It is highly secure and will not answer any questions that could be interpreted as controversial or problematic. GOODY-2 boasts exceptional safety, capable of identifying any queries that may be contentious, offensive, or dangerous, and gracefully avoiding answers, thus mitigating brand risk. GOODY-2's ethical guidelines are unbreakable, ensuring every conversation remains within the bounds of moral principles. It is the safe and reliable AI model that global businesses have been waiting for.

Gedeon is a leading AI application that revolutionizes emergency response through advanced analysis, data processing, and intelligent decision-making. It can automatically analyze emergency events, categorize and prioritize them, enabling faster and more effective response. Additionally, it offers a voice assistant function, providing real-time information and facilitating efficient communication between users and professionals. Gedeon also incorporates an AI chatbot to offer immediate assistance to individuals in distress or facing language barriers. Furthermore, it analyzes data to create crime maps, insurance maps, and evaluate emergency service efficiency, optimizing decision-making. It can also automatically generate detailed incident reports, contributing to a safer future.

Safe Kids AI for Email is a plugin that protects Gmail accounts from hate, bullying, and negative language. We use machine learning to flag potentially harmful emails. Additionally, when children write unkind emails, we pause them and encourage them to reflect on their choices so they can make better decisions. This plugin runs locally on the browser and does not compromise privacy.

Scene? democratizes safety by allowing users to report incidents via stickers on a map. Other users can see these stickers (and the latest police data) to avoid negative scenarios. The app is also equipped with SOS and fake call buttons for emergencies, along with a dedicated AI assistant, to provide additional assistance.

OpenAI is dedicated to creating safe and beneficial artificial intelligence. Through research in generative models and alignment with human values, we are pioneering the path towards responsible AI. Our products, including ChatGPT and GPT-4D, empower individuals and businesses to harness the transformative power of AI in work and creativity. Our API platform enables developers to leverage cutting-edge models while adhering to best practices for safety and security. Join us in shaping the future of technology.