There won’t actually be that many powerful AI models to choose from in the near future
The main players driving AI forward in 2023 might actually not look so different in the decades to come. Like much of the tech industry, generative AI could face a wave of consolidation and monopolization. Running and maintaining a ChatGPT or Google’s Bard, the authors write, is for now, prohibitively expensive.
The largest AI models, the authors estimate, cost around $1 billion each just to train. But the costs don’t stop there. Once the models have the necessary data, thousands of other computers running constantly are required to power the model and ensure it delivers “dick jokes in the style of Oscar Wild” to its users in a matter of seconds.
For now, the authors say it simply doesn’t make sense for most large companies, aside from Microsoft that is, to pay for exclusive use of those models. All of that means the creators of new large language models will likely turn to subscription business models for the foreseeable future, “so that a single model will serve the needs of many thousands of individuals and businesses.” In other words, while plenty of companies will hop on the generative AI bandwagon, only a handful will actually stand out.
