What is PCER?
The PCER System uses a variety of tools together to make an AI more efficient. Effectively, it is a workflow that combines…
The PCER System uses a variety of tools together to make an AI more efficient. Effectively, it is a workflow that combines the PCER Cycle (Prime, Craft, Expand, Refine), the concept of using tables (columns and rows in spreadsheets), and the process of prompt boarding (drafting several prompts to guide a task to completion).
Each aspect of the PCER System is valuable enough on its own, and you are welcome to it without buying my book. I made an .md file you can get with the preorder of the book that will streamline the process for you. Still, I imagine it is pretty easy to replicate because I derived it from my own research into how current LLMs (large language models) were developed and created. At the beginning of my research three years ago, all the companies were pitching basically the same product, with only minor tweaks based on whatever they could consider “unique” training data. They haven't evolved much in the interim. But the fundamental replication process of creating these models has led users to misunderstand their capabilities entirely. They are pattern-replicating models, not search engines. They are immensely powerful, but we use them misguidedly, seeking meaning and information from them rather than using them to organize the information we already have. Since we've been organizing our own information for centuries, their only niche and profitable human use is at the frontier: collecting and analyzing what we don't understand about the worlds around and inside us.
The use of AI to answer questions should only be when the agent is doing an internet search in your stead. The only art AI should create is measuring proportions and collecting references to inspire an artist. AI usage in this limited fashion would expand an individual’s horizons without the massive burn and blight of data centers on our environment. But does that replace a person in the workforce? Not necessarily; therefore, the Faustian economic bargain that has been made to bring about its existence is moot. Therefore, the PCER System is also a way to guide you through a creative or production process, so you can structure your ideas efficiently and pick yourself back up when this inevitably collapses.
Prime - Feed an AI agent with research articles and examples about and of your task
Craft - Create an outline of your project and the items you will need to build to complete the task
Expand - Add to each aspect of the project, exploring potential improvements to the original outline.
Refine - Cut out bloat from the project, getting rid of redundancies or rewriting based on the expanded outline.
TABLES - The AI agent can maximize its response and token usage if it organizes the output in columns and rows of information, reducing hallucinations or AI speak/jargon
Prompt boarding - I guess I like spreadsheets; we can present information in tables to an AI, keeping information organized and giving it a guide of prompts to follow - guiding an AI through the PCER Cycle and giving us an expanded product to refine, with research links we can double-check to prove authenticity. I would go step by step and reference a spreadsheet with my project outline of prompts.
Following this process should reduce your overall token burn, because you are not using the LLM agent mindlessly and getting inspiration as you go (changing your project scope). The user does prep before consulting the LLM, or uses the LLM to do prep work and look for peer-reviewed sources before developing unquestioningly. All processes, whether coding, marketing, or medical, should have a reference document for the LLM to follow, even if it was trained on that data. This is because all LLMs are pattern-replicating, and the pattern must be brought to the forefront for it to be truly useful.
Thanks for reading; preorder the book for more.