Orca LLM: The Possibilities of Semiotics and Design Thinking in Evolution and Advancement of Generative AI and Language Models.
The successful development of Orca illustrates the harmonious convergence of semiotics and technology, offering exciting opportunities for the application of design thinking methods in future LLM designs. Semiotic engineering, a discipline that considers interactive systems as meta-communication artefacts, emphasizes the importance of progressive semantization and the co-evolution of users and systems during usage. By integrating semiotic principles into the design process, future LLM designers can analyze the symbols, signs, and meaning behind user queries and responses, similar to how Orca interprets the reasoning process of GPT-4.
The rise of generative AI, particularly in the form of large language models (LLMs) like GPT, has sparked significant interest and concern within the design community. Designers are grappling with the implications of these technologies on creativity and society, leading to diverse reactions and questions about the future of design. However, rather than simply reacting to these advancements, designers have a crucial role to play in actively contributing to the development and improvement of AI systems. This article explores the integration of semiotics, design thinking, and meta-design as potential avenues for enhancing LLMs and fostering human-centred design in the context of generative AI.
The Semiotic Understanding of Orca
I am excited by the release of a research paper by the @Microsoft research team just a few days ago titled ‘Orca: Progressive Learning from Complex Explanation Traces of GPT-4’. This paper unveils the Orca model, a powerful system comprised of 13 billion parameters, which acquires a profound understanding of explanation traces, sequential cognitive processes, and intricate directives extracted from GPT-4. Orca leverages GPT-4 to learn explanation traces, sequential cognitive processes, and intricate directives. Orca’s integration significantly enhances the performance of existing instruction-tuned models and offers intriguing possibilities for the incorporation of design methodologies and semiotic knowledge. By studying and emulating GPT-4’s reasoning processes, Orca demonstrates a profound understanding of semiotics, the study of signs and symbols, which underlies the communication between users and systems. For example, by analyzing the prompts “think step by step” and “explain it to me like I’m five,” Orca not only learns from the answers but also from the reasoning exhibited by GPT-4.
The Convergence of Semiotics and Technology
The successful development of Orca illustrates the harmonious convergence of semiotics and technology, offering exciting opportunities for the application of design thinking methods in future LLM designs. Semiotic engineering, a discipline that considers interactive systems as meta-communication artefacts, emphasizes the importance of progressive semantization and the co-evolution of users and systems during usage. By integrating semiotic principles into the design process, future LLM designers can analyze the symbols, signs, and meaning behind user queries and responses, similar to how Orca interprets the reasoning process of GPT-4.
Enhancing Evaluation and Fine-tuning
Design thinking methods can also contribute to the evaluation and fine-tuning of LLMs. Orca’s extensive evaluation on multiple benchmarks showcases the power of rigorous assessment in measuring model performance. By incorporating design thinking principles, designers can develop comprehensive evaluation frameworks that go beyond simple question-answer pairs. This approach allows for detailed reasoning analysis and provides valuable insights into the strengths and weaknesses of the model, guiding iterative improvements and enabling LLMs to self-improve.
Meta-Design and User-System Interaction
The conceptual framework of meta-design offers another perspective for designers to bridge the communication gap between users and designers and facilitate better user-system interaction. Meta-design enables domain experts to create software artefacts tailored to the end users’ culture, skills, and background, empowering users to act as designers during system usage. By incorporating semiotic engineering principles, designers can shape interactive systems that promote continuous user-system co-evolution and address the challenges posed by generative AI and language models [1]. The application of meta-design has already shown promising results in various domains, including medicine, cultural heritage, assistive technologies, e-government, neuro-rehabilitation, and robotics.
The way ahead?
The convergence of semiotics, design thinking, and meta-design presents a promising path forward for the design community in harnessing the potential of generative AI and LLMs. By incorporating these methodologies, designers can shape AI systems that align with societal needs, values, and user expectations. This holistic approach to design not only enhances the performance of LLMs but also fosters sustainable and human-centric AI platforms.
Our response to generative AI and LLMs should extend beyond reactive measures. Designers have a significant role to play in contributing to the development and improvement of AI systems. By integrating semiotics, design thinking, and meta-design, designers can enhance user-system communication, support user co-design, and enable LLMs to self-improve. This proactive approach holds promise for creating AI systems that are not only technologically advanced but also aligned with human needs and aspirations.
For a friendly video introduction to Orca: https://www.youtube.com/watch?v=Dt_UNg7Mchg
Further technical reaching : https://arxiv.org/pdf/2301.13688.pdf, https://arxiv.org/pdf/2305.17126.pdf, https://www.semianalysis.com/p/google-we-have-no-moat-and-neither,
