AutoGPT (GPT3.5T) and ChatGPT (GPT4) - Scope of Work generation trials

1 Introduction

This article presents a comparison between two popular AI tools to generate a Scope of Work (SoW) document from limited specifications. The AI tools used in these examples are AutoGPT, which is based on the GPT3.5-Turbo architecture, and ChatGPT, which utilises the GPT4 architecture.

GPT4 and GPT3.5-Turbo (GPT3.5T) are both artificial intelligence models developed by OpenAI. They are instances of transformer-based language models, specifically built using the GPT (Generative Pretrained Transformer) architecture. These models are designed to generate human-like text based on the input they're given.

These trials were done to briefly show their utility in SoW generation and also to highlight what people non-technical-in-AI could achieve today and show the possibilities of what could come in the near future.?

2 Method

The intention of these trials is to provide the AI with very specific, factual elements of information that it will use to develop a Scope of Work document. The hope is, at some time in the future, that people can input specific 'needs' and 'outcomes' and the AI will utilise this information, along with other pre-known client data (such as site data and client standards) to form a context of the problem and develop a technically accurate SoW document.?

I have written very basic, dot point style client specifications, client standards, and site data for the AI to utilise in the SoW generation. In both cases of AutoGPT and ChatGPT I have used simple methods to read-in these three elements of information and have it confirm that it has read it.?Below are two decent examples of a trial of 9 different input data.

3 Input data

4 Outputs

4.1 ChatGPT (GPT4) Scope of Work document

The following is the raw output from ChatGPT (GPT4) after having read-in client specifications, client standards, and site data information.

4.2 AutoGPT (GPT3.5T) Scope of Work

The following is the raw output from AutoGPT (GPT3.5T) after having read-in client specifications, client standards, and site data information.

5 Findings, limitations, conclusions

• The examples demonstrate the potential horizons of Scope of Works development for future projects, offering an opportunity for corporate entities to innovate their methodologies.??
• The performance of ChatGPT (GPT4) significantly outshines that of AutoGPT (GPT3.5T), likely attributable to the superior computational power of the former. There is an increased contextual understanding within the outputs of ChatGPT, shown by its inference that should concrete be involved, it must be blue, and the pump's disposal subsequent to its removal. Furthermore, ChatGPT comprehends the necessity for suitable documentation as a deliverable in such work scopes.
• AutoGPT (GPT3.5T) offers an automated reading process post-prompt, enabling higher volumes of data to be parsed automatically. With solutions like Pinecone.io, static information, such as Standards, can be retained in AutoGPT's memory. Currently, the practicality of reading in comprehensive sets of standards with ChatGPT (GPT4) is limited. However, the feasibility of this approach warrants reassessment following the public release of the GPT4 API.
• Despite the accuracy of the assumptions made by ChatGPT (GPT4) within this limited dataset, an increase in the volume of input data may lead to erroneous assumptions, necessitating further scrutiny.
• It is important to acknowledge that the utilisation of these AI models does not remove the necessity for a thorough review and validation of the Scope of Works. Moreover, potential concerns arise from the transmission of potentially sensitive corporate data to an external AI "blackbox". Anticipating future trends, it is plausible that enterprises may transition towards proprietary AI solutions, affording greater internal control and security.