Summer is over. It's time to replace developers with AI and find out why MLOps is where DevOps and data scientists find a common ground.
The latest issue focused on NLP and machine capabilities to handle human language. In this new release, Autumn is coming with a bunch of great news, exciting papers, and unique service capabilities to be used by AI practitioners. In this issue, we discuss generative models for developers and automating all the things with MLOps.
Hot for coders
After the release of OpenAI GPT-3 last year, generative AI has been widely adopted. Many data scientists started experimenting with all the possibilities this could open, especially when dealing with the code that makes machines work.
Last month OpenAI launched Codex in private beta. Since then, several astonishing applications have been built in just a few weeks. Codex is a direct descendant of GPT-3, trained to support developers in code generation specifically. It is pretty similar to Github Autopilot but raises the bar of what can be achieved: check out the first demo and be blown off watching how it handles context and uses common concepts such as "what is visible on a background."
In the same domain, AWS just published content showing excellent features of Amazon CodeGuru: a video about CodeGuru Profiler and a blog article about Finding code inconsistencies using Amazon CodeGuru Reviewer
I don't think ML will replace developers, but it will be a great addition and speedup to everyday work, as stated in this article. Then, I am pleased to tell you GPT-4 is on its way to being released as a much smaller model and more tailored to specific applications.
Automating Machine Learning
Every DevOps engineer sooner or later gets confused (and a bit disappointed) by the seemingly lack of operational principles widely adopted in machine learning. The reason behind this is an entirely different approach adopted by data scientists that prefer being able to experiment at a swift pace instead of dealing with architecture.
This trade-off is something cloud vendors, and software architects are trying to solve with the new practice of Machine Learning Operations (MLOps, in short). In a Data Camp podcast episode, AWS Hero Noah Gift discusses a pragmatical approach to MLOps. At the same time, [AWS Community builder Alessandro Gaggia[(twitter.com/balubor) just published an article about how architects can apply software principles to machine learning.
More on this, if you have lived on Mars for the latest three years and just landed back on Earth, check out Julien Simon's show on Twitch SageMaker Fridays episodes. The latest of them is focused on pipeline automation. One week ago, Julien moved out from AWS to be the Chief Evangelist at HuggingFace. Congrats to Julien for his new role!
- re:Invent is back in presence and online for 2021. After a one-year online-only conference, people can now join the best cloud conference for AWS practitioners that the travel ban from Europe will be lifted in November. If already living in the US, you definitely should not miss this. Be ready to have your vaccine certification shown at the registration desk, to have a safe and funny conference.