Skip to content

About

Module 1 Challenge - Case Study on Company using AI

Resources

Stars

0 stars

Watchers

1 watching

Forks

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 

Repository files navigation

ai-case-study

Module 1 Challenge - Case Study on a Company using AI

ModelBest and ChatDev Innovation

Overview and Origin

ModelBest Inc. is a leading ai research companies in China. The company was founded in August 2022 with founding members from Tsingshua University in Beijing China. ModelBest seeks to innovate in the field of AI and expand what is possible with AI (ModelBest, n.d.).

ModelBest Inc. in association with Tsinghua University , the University of Sydney, and Beijing University of Poasts and Telecommunications (BUPT), created an opensource framework for using Large Language Models (LLM) to write software (Qian et al., 2023). Rather than using a single chat instance, they approached the problem of software development by splitting up the tasks amongst multiple chat agents. This approach more closely mimics the way that humans often develop software, as a team of individuals, each with different roles, collaboratively working together to write the software. The team used the waterfall software development method used in industry, albeit a method that is no longer popular.

The different chat agents were trained to specialize in different subtasks of the development and the natural language was used by each of the agents to communicate with each other.

Business Activities

At this stage, ChatDev is not successful enough to write complete software programs that are ready for market. In the future it may be, but even now the team believes that it is useful for brainstorming in software development. The goal is for the ChatDev to enable individuals and companies to realize their ideas for software even though they lack the skills and experience to do it themselves. Currently, AI can be useful in programming to offer coding suggestions, answer programming questions, and can aid in many other ways as well. However, at this stage, AI is just a programming aid and cannot develop software programs on its own. AI currently makes many coding mistakes and does not check its own work.AI currently commits what is known as hullicinations, meaning... ChatDev reduces many of these problems by subdividing tasks out to multiple agents which check each others work.

"The frequent occurrence of coding hallucination in turn reflects the constrained autonomy of agents in task completion, inevitably demanding additional manual intervention and thereby hindering the immediate usability and reliability of the generated software (Ji et al., 2023)" (Qian et al., 2023, p. 2).

Landscape

There have been many advancements in LLMs in the past few years with OpenAI leading many of those developments with the success of ChatGPT. LLMs have become far more accurate, relevant, sensical and ubiquitious in the last few years. They are now used as tutors, proofreaders, summarizers, planners, brainstormers, etc. Other major players are Meta, Google, Microsoft, and Apple which have all been incorporating these new advancements into their existing platforms like search engines, cell phones, social media, and more.

Results

ModelBest has created a model that is smaller than ChatGPT and yet performs very well. The smaller size means that the model is much cheaper to use. Recently, Stanford plagiarized ModelBests LLM becuase of its performance (Yuwei, 2024).

Recommendations

I would suggest that they try the chat agent approach with fields other than software development. This could lead to improved results in many fields that require collaborative effort Would LLMs be able to create more compelling movie scripts if the chat agents were split up into separate roles such as writer, director, storyboard artist, etc. The movie may also utilize image and video generating models to help bring new ideas into the industry.

About

Module 1 Challenge - Case Study on Company using AI

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors