Industry executives and experts share their predictions for 2024.  Read them in this 16th annual VMblog.com series exclusive.

How GenAI Will Transform Core Infrastructure in the Coming Year, From Custom Chip Development to Autonomous Driving

By John Hayes, CEO and founder, Ghost Autonomy

2023 was dominated by breakthrough after breakthrough in AI, perhaps none bigger than OpenAI's release of multi-modal LLMs (MLLMs). These new models are capable of understanding and drawing conclusions by combining video, images and sounds, opening up millions of new applications across nearly every industry. In 2024, on the back of MLLMs, I expect AI will dramatically expand its reach into all areas of technology, fundamentally changing how we write software and build hardware.

1. APIs will no longer speak in code, but instead in words and pictures

What we currently think of as APIs will soon become obsolete as LLMs usher in a new era of unstructured machine-to-machine communication. Traditional APIs and their reliance on structured data will cede dominance to natural language interfaces, enabling dynamic communication through text snippets and images. Businesses and developers will adopt LLM-powered interfaces that can interpret natural language and media inputs for more efficient, human-like interactions. 

Already available in applications like QR code scanning - where visual data is translated into meaningful actions - this approach ensures data transmission is streamlined and secure. Next gen APIs will leverage LLMs to understand complex queries that will provide truly seamless integration between applications, simplifying integration processes and enhancing user experiences.

2. AI supermodels will rapidly replace purpose-built models for millions of use cases

LLMs will turn the process of developing AI models inside out. Instead of relying on massive real-world data sets and human intuition, developers will be able to start with knowledge-rich models, fine-tuning them with a handful of samples for precise outputs. As these models become more intelligent, extensive fine-tuning becomes less necessary, making AI development more accessible and commoditized - you'll have to build fewer and fewer products customized to any particular model. 

While specialized models might still find niches in scenarios demanding specific performance or latency requirements, the trend is clear: eventually the current explosion of special-purpose models will consolidate into a "supermodel" with general intelligence that will be able to directly solve a wide array of really specific problems in AI. The rise of these supermodels will usher in an age where AI solutions are not just intelligent and deliver superior performance but are also economically viable and easier to build.

3. Move over data, LLMs will be the great equalizer in the AV race to full autonomy

Thanks to generative AI, data is no longer the new oil when it comes to autonomy. Until recently, the ultimate competitive advantage established players (read: Tesla) had was access to enormous data sets, sourced in the real world, for training their advanced AI and autonomous driving models. However, generative AI has erased this advantage by enabling anyone to create vast synthetic data sets at low cost, eliminating the requirement to collect so much real-world data to power models.

LLMs - with their fusion of general intelligence and specialized problem-solving capabilities - will serve as the great equalizer in the AV space. They will empower emerging companies to compete at a comparable level to the incumbents, fostering innovation and competition in the race towards fully autonomous vehicles. Generative AI has changed a critical part of the equation and is dismantling one of Tesla's biggest moats in the process.

4. The fervent pace of AI innovation will crush custom chip development

Developing custom chips in today's fast-paced environment is a misguided effort and locks companies into yesterday's innovations. Businesses that invest time and resources into making their own chips are unable to quickly adapt to unforeseen and sudden bursts of innovation. The most recent example of this is the advent of large language models (LLMs). All innovations - like the recent flurry of advances in LLMs - are always built on an open-standard industry leader. This won't change. Going forward, we'll see an industry-wide standardization on chips as companies fight to maintain competitive advantage amid the breakneck pace of innovation we're witnessing in the market. 

5. 2024 will see the rise of new data centers, optimized for large models 

In 2024, we'll increasingly see large models replacing the many smaller, purpose-built models that companies have previously relied on to create and train their AI models. As large models come to dominate across use cases in the coming years, new types of data centers optimized with compute at scale to run large models (i.e., kitted out with supercomputers) will replace traditional data centers outfitted with stacks of x86 machines.

##

ABOUT THE AUTHOR