Summary: Several industries require that Standard Operations Procedures (SOPs) are established critical business processes. With increasingly complex products and business processes, SOPs are becoming overwhelming for employees to follow. Large language models (e.g. ChatGPT) can be used effectively to guide projects and professionals through the jungle of process compliance. Large Language Models thus pose completely new opportunities for process-heavy companies such as Life science, Defense, Aerospace and Finance industries.
This article will outline:
- Why process-heavy organizations are challenged by a big landscape of process documents (SOPs)
- What role Large Language Models (like ChatGPT) might play in navigating process landscapes
- How to get started with training ChatGPT in your processes
The evergrowing complexity of the tech industry
We live in a world of increasingly complex products.
Remember your first cell phone? If you were born before the 90’s, your first cell phone were not much more than a landline made wireless by the use of a wireless transceiver, battery, controller, keypad, microphone and speaker. Relatively simple electronics even by the measure of its time of introduction.
Today’s smartphones are incredibly advanced and are as much capable sensor platforms as computing powerhouses. They must have – in addition to the above – GPS-receiver, WiFi, Bluetooth, NFC, accelerometer, gyroscope, magnetometer, ambient light sensor, multiple cameras (front and back), CPUs, GPUs, various memory components and of course, a display with capacitative touch interface, to just meet minimum expectations.
If you don’t think the iPhone 14 Pro above look that much more complex than the Nokia 3210, then consider this: In 1999 the fastest supercomputer at the time were the Intel Red/9632.
The Intel Red/9632 supercomputer was the peak of computing when introduced, performing computing workloads for defense purposes at an incredible 2.3 TFLOPS. Be so kind to turn it off when you’re done, because the electrical bill increases by a stunning 850 kw per hour.
If we consider the torn-down picture of a iPhone 14 Pro from before: The graphics processing units alone of its A16 chip is capable of… 2.0 TFLOPS. That is nearly as much as the supercomputer of 1999, but this one uses a mere 8 watts at peak. Fortunately, we do not have to turn off our smartphones before we go to sleep.
The burden of process-heavy industries
As products grow in complexity more teams will get involved in product development and manufacturing, and so, the processes of the organization also increase in complexity.
Tens of engineering divisions, and hundreds or even thousands of engineering teams, must work in careful orchestration to produce the first design of a smartphone today.
Process landscapes – of the corporations creating the products – grow exponentially to the complexity of the product. Why exponentially? Well, not just will the engineering teams increase in number; each team will also have to interface with more teams. As engineering teams become more specialized, detailed responsibilities must be defined, product governance must be designed and instructions specified.
For technology corporations that also operate in a regulated industry, such processes often become procedures. And that’s what we’re going to talk about!
Procedures come in the form of instruction-heavy documents that often must followed to such degree that an auditor can verify documented compliance. And guess what? Poor employees will have to read them as they get onboarded or the processes are updated.
Numerous organizations today have implemented, and operates by, a quality management system. This is even sometimes part of the license to operate within the Pharmaceutical (ISO 9001 + GxP), Medical Device (ISO 13485), Automotive (ISO 16949), Aerospace (AS9100) and Defense industries (AS9100).
Standard Operating Procedures (SOPs) hold a centerpiece role of Quality Management Systems (QMS) and specifies how activities must carried out (e.g. how to design software systems) and which controls must be carried out (e.g. how to perform cybersecurity reviews of said systems). Because SOPs often need to capture a large landscape of regulatory and stakeholder requirements – which must be expressed in the instructions – they have a tendency of becoming overly detailed with several convolutions and conditionals.
In defence of SOPs, they are very useful for ensuring consistency for well-defined, routine tasks such as sub-assembly manufacturing. However, when it comes to governing open-ended, project-based design and development activities, SOPs offers only vague guidance but imposes plenty of constraints. SOPs thus become distant in a day-to-days of projects, which risk re-work when the QA arrives to review the development activities of the team.
Quality Management Systems for the uninitiated
Numerous organizations today have implemented, and operates by, a quality management system. This is typically part of the license to operate within the Pharmaceutical, Medical Device, Automotive, Aerospace and Finance industries.
A Quality Management System (QMS) is an integrated set of policies and procedures designed to ensure that a company’s products or services consistently meet customer expectations – including regulatory requirements and other quality standards. Sometimes the implementation of a QMS is mandatory to operate in an industry. In other situations, organizations may simply evaluate that a QMS is an effective way to manage its operations (this is typically the case for CMMI and Lean Six Sigma management systems) or risks (the typical rationale for implementing an ISO 27001 Information Security Management System).
Producers of Medical Devices – as example – are all-but-required to implement a QMS that complies with the ISO 13485:2016 standard for design, development and manufacture of medical devices.
What can Large Language Models do to help?
Large Language Models – like ChatGPT – are exceptionally good at solving prompts in domains rich on natural language. SOPs are essentially domains rich in natural language, where the answer is buried.
huge amounts of natural language. SOPs are essentially
can take a prompt (e.g. a question) in a natural-language and generate a response in natural-language (or even, source code) based on its training data. GPT-4 – the newest version of ChatGPT – was basically trained on the public version of the Internet and hence, has a huge “knowledgebase” that it will interpret your prompt against.
Mind you though, that because the training data was basically the entire public version of the internet, it does not know the specifics of how your company operates. For this, we will need to fine tune – which basically means, that we will provide additional training with which it can interpret questions.
How to train ChatGPT on your company knowledgebase?
Link here: https://beebom.com/how-train-ai-chatbot-custom-knowledge-base-chatgpt-api/
Link here to OpenAI GitHub: https://github.com/openai/openai-cookbook/blob/main/examples/Question_answering_using_embeddings.ipynb
Using ChatGPT for helping SOPs
- Ask ChatGPT when performing
Leave a Reply