Amazon Web Services Inc. recently made an announcement regarding their commitment to responsible generative artificial intelligence during their annual conference in New York. As part of this commitment, new capabilities like Contextual Grounding Check in Guardrails for Amazon Bedrock were introduced to help customers detect hallucinations in model responses. This new feature can reportedly filter out over 75% of hallucinated responses for various workloads.
The team at AWS explained that hallucinations can occur in generative AI models due to incorrect predictions of the next best token or word, leading to inaccurate information being generated. To address this, Amazon is also focusing on relevance, ensuring that the model responds appropriately to user queries.
In addition to Contextual Grounding Check, AWS announced new tools and capabilities for safe generative AI, including Guardrails for Amazon Bedrock and Evaluating models on Amazon Bedrock. These tools aim to introduce safeguards, prevent harmful content, and evaluate models based on safety and accuracy criteria.
Amazon highlighted their dedication to testing and teamwork for AI systems, using methods like manual team evaluation to ensure the security, privacy, veracity, and fairness of their models. They also introduced watermarking features for AI-generated content to combat misinformation and increase transparency.
Collaboration with organizations like the US Artificial Intelligence Security Institute Consortium and Thorn is helping Amazon develop trustworthy AI and prevent the misuse of generative AI technologies. Transparency is also a priority for Amazon, with the launch of AI service cards to provide insight into responsible AI design choices and best practices.
Amazon’s commitment to the safe and responsible development of AI extends to society and education, with initiatives like Brainbox AI aiming to optimize energy use and reduce carbon emissions. The company is also providing training on risk assessments and responsible AI use for employees, as well as launching courses for the public through their “AI Ready” initiative.
Overall, Amazon continues to lead in creating a safe and trustworthy environment for AI development, with a focus on customer success and accuracy. Their efforts to address challenges like hallucinations and misinformation in generative AI demonstrate their commitment to responsible innovation in the field.
Article Source
https://siliconangle.com/2024/07/10/amazon-updates-commitment-responsible-use-generative-ai/
