AI companies that have spent billions of dollars building large language models (LLMs) are now banking on smaller language models as a new source of revenue growth, the Financial Times reports. The costs of creating and training such AI models are much lower than for LLMs.
Apple, Microsoft, Meta, and Google have recently released new AI models with fewer parameters but still powerful capabilities. They were prompted to take this step by the high cost of LLMs, as well as concerns about the use of data collected for training due to possible copyright infringement.
Advantages of Smaller Models
Companies like Meta and Google have begun offering small language models with just a few billion parameters as a cheaper, power-efficient, customizable alternative to large language models. In addition, such models require less energy to train and run, and they can protect sensitive data.
“By having that much quality at a lower cost, you’re actually giving clients the ability to use a lot more applications and do things that they thought wouldn’t have given them enough return on that (LLM) investment to justify actually using it,” said Eric Boyd, corporate vice president of Microsoft Azure AI Platform.
Google, Meta, Microsoft, and French startup Mistral have also released smaller language models that are more advanced and task-oriented. These models have the advantage of performing tasks locally on the device without sending information to the cloud. This will appeal to privacy-conscious customers who don’t want to send information outside of internal networks, notes NIX Solutions. We’ll keep you updated on further developments.
Local AI Processing and Future Developments
Small language models can also be used on smartphones. For example, Google’s Gemini Nano model is installed on Pixel and Samsung S24 smartphones. Apple has hinted that it is developing AI models to run on iPhone smartphones.
In turn, OpenAI CEO Sam Altman said his company will continue to work on creating larger AI models with advanced capabilities that can reason, plan, and execute tasks, which will eventually be able to reach the same level of intelligence as a person.