NikCreate

I recently attended AWS’s Global Summit in New York City and don’t remember an AWS event with such a clear theme. As with the world in general, Generative AI was the focus everywhere from talks on prompt engineering, many existing vendors advertising their app’s new AI functionality and the keynote itself. Many new products and features were announced, but all were related to building your own AI via their Bedrock service, vector database support or LLM integrations with their other services.

The conference left a weird impression. It was a large and well-run event by all means, but Generative AI’s shadow left me unsure how to feel. Like the rest of the world, I’ve been trying to figure out how to think about the rise of ChatGPT. I have played around with it and a variety of Microsoft and Google’s integrations along with Midjourney (for blog post images) and a bit of Langchain to ask questions of my documents. I’d heard of the GPT models prior to the ChatGPT craze, but since my Kaggle blogs a few years back, I’d mostly shifted my focus to going deep into AWS and Cloud Infrastructure and hadn’t done much with AI besides playing around with DeepRacer a couple months.

While Generative AI is so popular, it can be exhausting to hear about at times, there’s a lot of potential around it where it seemingly can’t be ignored. Its text and code writing ability at present shifts between scary, at how fast it can generate some code or explanations. At least when it isn’t hallucinating. While learning to work with it better is definitely a skill I’ll need to master, It’s potential to grow from here makes me regret not spending more time on ML/AI in my last couple years to help push the technology forward. And of course, the incorporation of agents, allowing the AI to take action seems like a world changing application in itself.

Similarly, I have a couple different takes on this conference. Having paid attention to advancements in generative AI and its integrations, everything announced and discussed seemed to lag behind what was already available. For instance, Bedrock’s introduction of Agents is definitely a necessity, though something already available via ChatGPT, LangChain, AutoGPT, etc. The Keynote and a lot of the conference ended up feeling a corporate rehash of a viral “ChatGPT Productivity Hacks” YouTube videos, such as Salesforce enthusiastically demonstrating their ability to generate emails for customers.

Of course, unlike Microsoft and Google’s more public facing attempts to one up each other, the real draw of AWS’s announcements is that customers can utilize Generative AI within AWS. A big hurdle that stops companies from getting more out of AI is trust concerns, since companies can’t just send their code bases, data or secrets to ChatGPT. On the otherhand, if they already trust AWS with their data, these announcements would help them utilize AI without the same security issues. AWS offers tools for helping generate models with knowledge from their data sources, which is easier than spinning up an LLM from scratch, and the ability to invoke AWS actions via their agents, which seems like a natural fit for Lambda.

Having just finished renewing the Solutions Architect Professional certification, it’s interesting to see a group of services and features which will likely permeate much of the AWS console. It will be interesting to see how use of these tools changes the current way we interact with AWS in the future. With certifications on the mind, it will also be interesting to see what happens to the current Machine Learning Specialty which was created prior to the Generative AI craze.

As for my next steps post-conference, I want to better familiarize myself with the workings of Generative AI and see how it compared to the Machine Learning tasks I’d done in the past.