Imagine a world where machines possess true intelligence, understanding the natural world and reasoning about it. We’re not quite there yet with current AI systems, which are limited to linking data points and reporting conclusions. But we’re on the cusp of a new era of deep learning and automated discovery that goes beyond simple language models.
According to Yann LeCun, AI Chief at Meta, current AI systems are “not even as smart as a dog.” While they have limited reasoning abilities, they lack a deep mental model of the underlying reality. However, the potential for AI to unlock brand-new knowledge is starting to emerge.
Take AlphaDev, the AI system built by Google DeepMind, for example. It recently discovered sorting algorithms that are significantly faster than those developed by data scientists and engineers over decades. Computers are now uncovering things that we don’t know.
We’re entering the age of AI-automated discovery, where machines can become domain specialists in minutes or seconds, advancing our understanding and making discoveries at lightning speed. This requires a new approach to AI infrastructure and a central corpus of the world’s data.
Unlocking brand-new knowledge at lightning speed
With thousands of data centers worldwide, each with massive computing power and vast amounts of data, we can create a global computer that accelerates the learning process. This data-driven and event-driven automation expedites AI discovery, impacting all of humanity and opening up new possibilities in uncharted areas.
“Imagine these machines tackling crop production, new approaches to sustainable energy, the elimination of disease,” says Jeff Denworth, co-founder at VAST Data. “We think that these machines will find and discover whole new domains of science and mathematics on their own that are beyond our current evolutionary step.”
But to make this happen, we need a new way of approaching AI infrastructure. We must build a thinking machine that can understand and process the rich natural data from the world. This requires a shift in data platforms and a focus on unstructured data.
Building a central corpus of the world’s data
To achieve AI-automated discovery, we need a next-generation approach to data management and database architecture. Today’s data management constructs struggle to handle the diverse types of data that AI initiatives ingest. VAST Data is simplifying the data management experience by breaking down infrastructure tradeoffs and rethinking the relationship between structured and unstructured data.
“Unstructured data, GPUs, global data sets, a variety of on–premises and cloud computers — these are the hallmarks of the environments that are being deployed by leaders in deep learning,” says Denworth. “With a full rethink, we can democratize systems of AI-automated discovery for everyone.”
For a deep dive into VAST’s vision for the future of AI infrastructure, plus a look at how customers like Pixar, Zoom and The Allen Institute and partners like NVIDIA are harnessing this powerful new approach to deep learning, don’t miss VAST’s Build Beyond event on August 1st.
For nearly a decade, the tech industry has been driven by advances in artificial intelligence (AI) and natural language processing (NLP). We have seen dramatic progress in tasks ranging from object recognition to automated translation and question-answering. Much of this progress can be attributed to advances in the development of large-scale language models (LLMs). These powerful, computationally-intensive models are now the driving force behind much of the AI revolution.
But as LLMs become more sophisticated and powerful, they are increasingly taking up immense amounts of energy and computing power. This, in turn, has triggered a pressing need to create the next wave of computing: algorithms and models that remain as powerful as LLMs but require far fewer resources to operate.
At the core of this challenge lies the task of making these new models truly generalizable. This means developing computer-learning algorithms that are not only proficient in the traditional NLP tasks, but also able to transfer knowledge across different contexts and languages. In essence, such models must be able to learn from a single language in order to navigate and assess other languages.
One approach to this challenge has been the development of generative pre-training models. Such models are designed to represent the underlying semantic structure of a given language and then use this structure to “fill in the blanks” for tasks like question-answering.
Such models have already shown promise in languages such as French and German. In the near future, it is likely that such models will be extended to multiple languages including Japanese, Chinese, and Spanish.
At the same time, researchers are also exploring the potential of so-called “synthetic” models, which are specifically designed to operate across multiple languages. Such models combine elements from several languages into single architectures, allowing them to better capture the subtle nuances between languages.
Finally, advances in quantum computing are driving the development of even more powerful and generalizable models. By harnessing the unique features of quantum computing, researchers are pushing the boundaries of what can be achieved in terms of computational power and accuracy.
Overall, the tech industry is on the cusp of ushering in a new wave of computing – one that is driven by more sophisticated models that are able to generalize across multiple languages and contexts. It goes without saying that such a breakthrough will have major implications for the future of AI and the NLP industry.
