Large Language Models are the new trend for all good reasons. These models use deep learning techniques and are trained on large amounts of textual data. They produce human-like text and perform various Natural Language Processing (NLP) and Natural Language Understanding (NLU) tasks. Some famous LLMs like GPT 3.5, GPT 4, BERT, DALL-E, and T5 are performing various tasks like generating meaningful responses to questions, text summarization, translations, text-to-text transformation, and so on.
Recently, a new approach called dreamGPT has been introduced, which makes use of the power of hallucinations from Large Language Models to stimulate divergent thinking. This innovative approach helps in generating unique and creative ideas. While on the one hand, where hallucinations are typically associated with a negative connotation and are mostly referred to as a drawback of LLMs, DreamGPT enables the transformation of hallucinations into something valuable for generating innovative solutions.
The current LLMs are mainly designed to address particular problems by understanding and generating text based on instructions or prompts. But, these models are limited to generating responses that align with existing patterns that they have learned from the data they have been trained upon. This restricts their ability to explore alternative or unconventional ideas. Here comes DreamGPT, with a different methodology to make use of the inherent capacity of LLMs to hallucinate. During the generation of text, the production of a text that may not have a direct basis in reality but can still be useful and creative is the aim of this approach.
This can help dreamGPT explore different use cases and use divergent thinking. Divergent thinking refers to generating a wide range of creative ideas, considering multiple perspectives, and exploring different solutions. Using this, DreamGPT can explore as many possibilities as possible instead of just aiming for a single correct answer or a specific problem-solving approach.
To use dreamGPT, the users need to install Python 3.10+ and Poetry. Poetry is a tool that is used for dependency management and packaging in Python. It allows the declaration of the used libraries in a project and helps in installing and updating them. DreamGPT works in a loop by planting random seeds, dreaming about new and creative ideas, combining and evaluating different approaches, selecting the most novel approach, and repeating it in a cycle.
dreamGPT is open-source in nature and can run locally on any PC or Mac without the requirement of a GPU on the device. Samples have been shown on Github’s Readme file, which can be accessed at here. On running DreamGPT, it generates a random seed of concepts and uses them as a starting point for its dreaming process. Each idea is evaluated based on diverse criteria, and the score is used to reward the best ideas over time. With the growth in population, the results improve.
In conclusion, dreamGPT is a great approach that embraces the hallucinatory capabilities of LLMs and seems promising for stimulating divergent thinking and generating innovative ideas.