Season 4 Episode 15: Models, Money, and Monitoring

In Season 4, Episode 15 Karl & Jon are joined by AWS Community Builder and core member of the Armenia UG*, Tigran Gevorgyan. They discuss Meta's Llama 4 models on Amazon SageMaker, prompting for best price performance with AWS AI models, patched vulnerability in the Amazon EC2 SSM agent, using OpenSearch and Managed Grafana for telemetry data correlation, updates to the AWS Well-Architected Framework and then the guys started discussing the merits of different sports teams' merchandise, completely forgetting they were supposed to be talking about AWS news.

 

09:02 - Meta's Llama 4 models on Amazon SageMaker JumpStart 

The article discusses the availability of Meta's Llama 4 models on Amazon SageMaker JumpStart. The speakers highlight the importance of model naming conventions and note that SageMaker is getting new models, which is significant given the recent focus on Bedrock. They discuss the differences between Bedrock and SageMaker, with SageMaker offering more control and customization options.

15:26 -  Prompting for best price performance with AWS AI models 

This article from the AWS Machine Learning blog focuses on optimizing price performance through efficient prompting. The discussion compares the costs of OpenAI's GPT models with Amazon's Nova models, highlighting potential cost savings. The speakers also touch on the differences in prompting techniques between the two systems and the potential challenges of transitioning from GPT to Nova.

22:37 - Amazon EC2 SSM agent vulnerability patched 

The article discusses a recently patched vulnerability in the Amazon EC2 SSM agent. The speakers explain the nature of the vulnerability, which involved privilege escalation via path traversal. They emphasize the importance of keeping systems updated and note that while the vulnerability required multiple steps to exploit, it's crucial to address such issues promptly.

28:29 - Correlating telemetry data with Amazon OpenSearch Service and Amazon Managed Grafana

This article from the AWS Big Data blog discusses using OpenSearch and Managed Grafana for telemetry data correlation. The speakers debate the complexity and cost of this solution, particularly for Kubernetes environments. They suggest considering alternatives like AWS X-Ray or Datadog, depending on specific needs and budget constraints.

34:53 - New guidance in AWS Well-Architected Tool 

The podcast discusses recent updates to the AWS Well-Architected Framework, including 78 new best practices. The speakers emphasize the importance of staying updated with these best practices and encourage solution architects and DevOps professionals to utilize this resource. They also criticize the pricing structure of the Well-Architected Tool, which is tied to AWS support subscription levels.

 

*AWS Armenia user group (https://www.linkedin.com/company/aws-user-group-armenia/), and community day (https://aws-community-day.am/)