AI Is Here. The Dawn of the Deep Learning Era
“AI is here.”
When we say that:
- Technical people sometimes ask why would we proclaim that phrase when AI has been around for decades? Of course it’s here.
- Business people usually say that statement is so true – have you seen the economic impact that is being created by AI, let alone the expectations for the impact it will have, as they are now seeing real business value derived from it
- At SambaNova, we say, of course it’s here, it’s in our DNA and we feel passionate about driving its marketplace success.
As AI has been around for a long time, some might even say it has been around since 1956. To a degree, that is true, but we have now reached the point where everything has changed. We are at the dawn of the deep learning era. So what leads industry experts to now say that AI will have profound impacts in the 21st Century, like the Internet did for the 20th Century, and Electricity for the 19th Century?
Whenever any technology reaches an inflection point where it displaces previous ones, it is usually due to a confluence of factors. In the case of AI, we have reached an inflection point as a result of:
- The existence of massive data sets. Deep learning models require vast amounts of data to train on to achieve tolerable levels of accuracy.
- The ubiquity of unstructured data. Previous applications relied on structured data, such as text and numbers, but the overwhelming majority of data created today is unstructured data. This includes text, sound, images, documents, videos, presentations, and everything else that constitutes the majority of data created today. The ability to process and analyze this data is a requirement going forward.
- Software applications have evolved. Software 1.0 systems, including those for transaction processing, ERP, core banking, taxation, and human capital management were deterministic computations written by experts with domain knowledge of every possible outcome as sequential, imperative code. Now, software 2.0 has arrived where the applications are instead developed by training data input into a probabilistic model encoded in training weights, which requires a Dataflow architecture.
- Availability of next generation infrastructure. The infrastructure that was built for software 1.0 has been rendered inefficient for software 2.0. Now, infrastructure that is purpose-built for Dataflow applications is available, both for on-premises applications and using an as-a-Service model to simplify adoption and implementation.
- Models have become significantly more sophisticated. For example, five years ago large language models were used for translation. Now, they have advanced to the point where in addition to translation, they can understand context and have generative abilities. They can now generate text and documents. Beyond even that, they can produce speech and even images that rival what human artists can do. The ability to produce content is the most remarkable aspect of this technology.
- The accessibility of models has changed. Just a few years ago, if an organization wanted to do deep learning they would have needed to hire a team of PhDs that would be working on developing the next model. Now the models are proliferating at a much lower rate and it is the utility of Transformer-based models that is driving this. The models have become more capable, they now require less customization, and they can be accessed and consumed as a service.
Beyond breaking the accessibility barrier and the evolving maturity of deep learning models and Dataflow computing, the other key factor that has changed is how AI is being delivered. Today, AI is available with complementary hardware and software, provided as a complete solution, which addresses the problems associated with the required inputs of massive volumes of structured and unstructured data that really brings this all together. Using pre-trained models available as-a-Service, these solutions enable end users to get deep learning projects up and running quickly without the complexities of self-integration and trial and error.
There are a volume of companies actively using AI in a material way. These are not niche organizations with small AI implementations. These are large enterprises that are moving markets with AI. AI is a market moving force today.
These marketplace realities are why AI is here.
Check back to read the next blog in the series and discover how leading organizations are overcoming the deep learning deployment gap.