Industry executives and experts share their predictions for 2024. Read them in this 16th annual VMblog.com series exclusive.
Real-World AI Utilization, Kubernetes Platform Toolkits
By Shaun O'Meara and John Jainschigg,
Mirantis
2023 was the year of the public large language
models (LLM) - with many companies experimenting and launching products that
rely heavily on these public models for a wide range of use cases ranging from
extremely complex to the very simple.
We believe that in 2024 we are going to see a
number of things change in the AI space as real-world adoption runs into the
realities of how to implement AI that's fit for purpose, secure, and private.
- More people will be looking for ways to run AI
models locally that are not shared and can be used to process private data
or create user interactions in sensitive environments (e.g. health,
finance). This will lead to technology companies providing simpler, more
"cookie cutter" approaches to running AI models and their dependent
systems.
- The simplification and commoditization of AI
models through the use of RAG (Retrieval Augmented Generation) allows
organizations to use pre-trained models and quickly provide value without
the need to invest in training of AI models. This, combined with point 1
above, will lead to technology companies providing out of the box RAG
solutions and a market for pre-trained models that are more specific to
RAG use cases and interactions.
- More specialized "point solution" AI and ML
models will emerge that can run on lightweight commodity hardware,
providing for very specific intelligence needs, essentially AI on IOT. The
reality is that not all AI functions need LLMs, but the LLM approach of
shared and commoditized models can be extended to the IOT. This will also
help reduce resource overheads in terms of power, cooling and hardware.
Standardized open source
Kubernetes platform engineering toolkits
As we predicted last year in VMblog, the Kubernetes
battlefield has indeed shifted away from the raw platform (now effectively a
commodity) and up into platform engineering.
This year, we expect to see lots of motion
around providing customers with ‘actually usable' production Kubernetes
platforms - collections of integrated services that equip a customer to develop
and test software, quickly build standardized and shareable automation, handle
security and access control (and compliance) in standardized ways.
The broadest change that we expect to see in
2024 from companies that have implemented Kubernetes platforms in production is
a move to real multi-cluster/multi-cloud deployments based on consuming
multiple different infrastructure providers (public cloud vendors and
on-premise) focused on treating the underlying Kubernetes clusters as
commodities that can be switched out as needed. This change will lead to more
standardization of the underlying Kubernetes clusters and a greater focus on
the tools and capabilities needed to create consistent application environments
for developers. And this, in turn, will lead to a focus on building effective
platform engineering practices and the software solutions to support them.
Kubernetes has become the de facto standard
for creation of new applications, but it is still complex to use and has a
steep learning curve. As a result of
this complexity and the drive towards multi-cluster application platforms, we
expect to see more tooling that helps developers abstract from the complexity
of Kubernetes and increase developer productivity.
Operational complexity and ML Ops
The rise in complexity of information
technology solutions has grown alongside our general dependency on these
systems in all aspects of daily life - this at a time when it is getting harder
to find and retain the specialized skills needed to maintain these systems.
This is not a new problem, and organizations have long attempted to address
these challenges through greater levels of automation. But automation also
requires specialized skills.
AI leaps over the last 18 months have led to a
situation where it is becoming more feasible to utilize AI and ML tooling to
create operational tools that can augment and reduce the need for the number of
skilled resources - the AI helping developers both in building reliable
automation "on the fly" (another example of AI-assisted coding) and, more
impactfully, in helping analyze metrics and observability data, do forensics to
determine root causes of issues, and (eventually) to direct, participate in, and
action mitigations and other operational processes (i.e., "OpsGPT: update my
cluster"). We expect that in 2023 we will see a greater number of ML and AI
Operations solutions being brought to market as well as realistic open source
solutions.
##
ABOUT THE AUTHORS
Shaun O'Meara is global field CTO at
Mirantis and has worked with customers designing and building enterprise IT
infrastructure for 20 years.
John Jainschigg is director, open source
initiatives at Mirantis, and is a cloud engineer, software developer,
content/product marketer, and technology journalist.