Meta develops an AI language tool: Toolformer

Toolformer is a language model designed by Meta researchers. The language model is unique since it has the capacity to utilize external sources. It is built on a pre-trained GPT-J model. It has almost 6.7 billion parameters. Researchers carried out various testing and experimentation with the Toolformer language model. Toolformer showed strong performance. The results indicate that Toolformer has the capacity to outperform GPT-3 model. Since GPT-3 model is based on only 175 billion parameters.

Well, this is not the first time that researchers have tried to present solutions to the limitations of language models. Just recently it was declared that Bing Chat has the capacity to perform web searches on its own when required. Where several other attempts have been made to incorporate it with calculators, search engines, and browsers.

In accordance with the Meta researchers, the current approaches that consider the integration of tools in the language models rely on a huge number of human annotations. Or in another case, they are restricted by particular task-specific settings.in comparison, Toolformer can learn to use a variety of tools in a more generalized way. Since it doesn’t need specialized training for particular tasks.

Given this perspective, it can be anticipated that in the future LLMs presented with the capacity to utilize external sources will become more reliable and trustworthy assistants. It would be due to the similar techniques utilized in Toolformer. On the other hand, the capacity to perform API calls can also have certain disadvantages. Since the user data saved in apps could be lost. Or it could pose a threat to the outside world via a web browser or communication-based tools. It has the potential to accidentally induce while providing an answer.