Introduction

​

GenAISafety, a global leader in transforming workplace health and safety (HSE) through generative artificial intelligence, combines advanced technologies to provide innovative and secure solutions. This explanatory note outlines the use of OpenAI technologies as part of our corporate membership, along with the integration of open-source models and benchmarks for security and performance.

Use of OpenAI Technologies

​

GenAISafety has established a corporate membership with OpenAI, enabling the use of their cutting-edge technologies under the highest security standards.

​

Key highlights of our use of OpenAI technologies include:

• Data Security: We exclusively use non-sensitive data for training and developing our models, ensuring the confidentiality and security of information.

• Compliance: All operations comply with OpenAI’s security regulations and guidelines, ensuring responsible and ethical use of generative AI.

• Advantages of Generative AI: OpenAI models, such as GPT-4, are integrated into our platform to analyze real-time data, anticipate risks, and provide personalized recommendations to enhance workplace safety.

​

Integration of Open-Source Models

In addition to leveraging OpenAI technologies, GenAISafety integrates open-source models to diversify and strengthen its analytical and predictive capabilities.

​

Key Open-Source Foundation Models:

​

• Bloom: A multilingual language model developed by Hugging Face, utilized for language comprehension and generation tasks.

• BERT: A pre-trained model by Google, applied to natural language processing tasks.

• MISTRAL 7B: A French language model developed by INRIA, designed for specialized applications in French.

• FALCON 180: A model tailored for advanced language processing tasks.

 

Security and Performance Benchmarks:

​

• MMLU (Multimodal Language Understanding): Evaluates multimodal comprehension of models.

• ACR (Automatic Content Recognition): Assesses the capability for automatic content recognition.

• BLEUScore: Measures the quality of automated translation.

• HELM (Human Evaluation of Language Models): Gauges model performance through human evaluations.

• TRIVIAQA: Tests question-answering capabilities.

• LMSYS (Language Model System): Evaluates overall performance of language model systems.

 

Purpose of Product Demos

​

The product demos available on [GenAISafety Channels] serve as illustrative examples of our solutions. These demos aim to showcase the capabilities and benefits of our products without involving sensitive or personal data.

 

Compliance with GDPR and Quebec’s Law 25

GenAISafety strictly adheres to the European Union’s General Data Protection Regulation (GDPR) and Quebec’s Law 25 on the protection of personal information.

​

• Personal Data Protection: We do not process personal data in demos unless explicitly necessary and with informed consent. We respect individuals’ data protection rights as defined by GDPR and Law 25.

• Transparency and Information: We clearly inform users about the purpose of data processing, the legal basis, recipients, and retention periods, in accordance with GDPR and Law 25 requirements.

• Data Security: Appropriate security measures are in place to protect data, including conducting data protection impact assessments when required. These measures comply with Article 32 of the GDPR and Law 25 standards.

• Incident Notification: Any confidentiality breach involving personal information is promptly reported to Quebec’s Commission d’Accès à l’Information (CAI) and affected users, as per Law 25.

• Consent and Oversight: We obtain clear and informed consent before collecting or using sensitive data. A designated Data Protection Officer (DPO) ensures compliance.

 

GenAISafety ensures compliance with Quebec’s Law 25 in its product demos by adhering to key requirements and principles outlined in the legislation.