Lionel Sujay Vailshery
Research expert covering cloud computing and emerging tech industries
Large language models (LLMs) - statistics & facts
While the term might sound like particularly verbose runway walkers, large language models (LLMs) are computational models whose aim is to achieve language generation, that is the use of language, in a natural form. That is using language like humans use it, accounting for nuances and the difficulties of human interactions. In simpler terms, LLMs are sophisticated software programs capable of classifying and using language without the direct commands of a human operator. The most popular of the LLMs in 2024 comes in the form of generative AI. This has drawn in the largest tech giants and the most advanced of AI startups in a revolution impacting across the globe.
The reason funding is so vital is that most AI startups are not profitable. This comes from many factors: a growing market, high energy and material costs when running the models, and the time and skill needed to develop them. Because of this, the competition is lopsided in favor of the large tech giants, as they have the billions to spend to improve the LLMs they run.
A large field dominated by a few giants
A variety of LLMs are available on the market, coming in multiple forms and offering a number of different capabilities. They are generally divided into open-source and closed-source, that is, those where the underlying code is freely available to the public versus those where the makers keep a tight lid on anything behind the scenes. While the market is diverse, there are a few models that are leagues ahead of others. These are ChatGPT, Gemini, Claude, Llama, and Mistral. Across AI benchmarks, these models consistently rank the highest, being capable of solving various tasks faster and more accurately than others.Funding plays a huge role
One of the reasons these models do so well is that each is supported by some of the largest and wealthiest tech companies on the planet. Google, Microsoft, Meta, Amazon, Apple, and Nvidia each have a stake to some degree in the success of these generative AI models, either through ownership or long-standing cooperation agreements. Mistral, developed by the company with the same name, is something of an outlier. It is the only European offering and focuses on a smaller model with less economic heft behind it. However, it formed a partnership in 2024 with Nvidia, a vital part of its continued well-doing.The reason funding is so vital is that most AI startups are not profitable. This comes from many factors: a growing market, high energy and material costs when running the models, and the time and skill needed to develop them. Because of this, the competition is lopsided in favor of the large tech giants, as they have the billions to spend to improve the LLMs they run.
LLMs growing larger
The need for AI capabilities is unlikely to decrease, meaning these models will grow even larger, requiring more energy, water, and labor to develop. This will necessitate more money and more precise models, reducing the resources needed to run them. This is beyond the capacity of smaller startups, and so it is likely that we will see fewer models run by larger corporations, taking over startups as they mature.
