False Steps: Coop’s Guide to Corporate Fallacies
Or: What LLMs still Cannot Solve
Good morning, good evening, good tidings and greetings
And welcome, my friends, to these whimsical readings
Despite the potential of algos and agents,
The apps we can conjure with minimal patience,
The tools we deploy with their untapped potential,
A human’s perspective can still be essential,
And so ’twixt the quipping and rhyme-laden twaddle,
Lies insight unseen from your large language model.
There are still broken processes that neither GPT, nor Claude, nor all the king’s LLMs and all the king’s AI researchers can put back together again. (Note: If, by the time you read these words, that statement is false, feel free to ask your intellectually-superior overlords for another form of entertainment or sustenance and return to whatever UBI-supported experience modernity has afforded you).
All dystopian snark aside, the capacity of agents to generate prose, code, and imagery is staggering and improving on a daily basis. Where once, entire teams of young professionals were required to pen marketing copy, visualize data, and develop all manner of software, these problems are quickly becoming the purview of the algorithm. Often, it is convenient, even rational to ask one’s friendly companion AI to answer the quantitative questions that arise in the course of a corporate professional’s day. I suppose, on some level, this is enormously cheap when compared with the cost of well-dressed professionals tasked with “business intelligence” or “advanced analytics” or “data science.”
The challenge is that all the insight a well-trained transformer can deliver is still limited by the human beings who turn that insight into decisions. As much as this bookish, neurodivergent corporate professional would wish it otherwise, life is not a math problem. Even after rational discourse and rigorous calculation, we still make a rather predictable set of poor choices. And while the ubiquity of agents might replace the army of spreadsheet jockeys and the middle managers who supervise them, the same human frailties will remain. Only this time, we’ll be working with attention spans ruined by infinite scrolling and tolerance for frustration long atrophied by use of large language models to ameliorate every intellectual struggle.
This book offers no solutions to the perils of doomscrolling nor the widespread diminution of perseverance. But it does offer concrete examples of how we are led astray by not simply data, but by the nature of human beings and the organizations that employ them. Also, in many cases, it rhymes. No reason this can’t be fun.
Chapters
The book proceeds through the ways sound analysis goes wrong — each chapter a false something, a different error of quantitative reasoning dressed in a corporate suit.
| Chapter | The error it names |
|---|---|
| Introduction: Dear C{x}O | the four blindspots of the data-driven executive |
| False Start | starting projects too soon |
| False Positives | saying “yes” when we should say “no” |
| False Alarms (in progress) | saying “no” when we should say “yes” |
| False Positivity (forthcoming) | believing the unbelievable |
| False Profits (forthcoming) | why every projection is optimistic |
| False Flags (forthcoming) | assigning credit or blame to the wrong party |
| False Identity (forthcoming) | why we cannot see ourselves clearly |
| False Witness (forthcoming) | why we cannot see others clearly |
| False Pretenses (forthcoming) | why we cannot say what we actually think |
| False Advertising (forthcoming) | presenting ourselves dishonestly |
| False Consciousness (forthcoming) | how ignorance of the basics dooms us early |
| False Equivalence (forthcoming) | why experimental results never equal reality |
| False Hope (forthcoming) | why what we believe will save us never does |
| False Accusations (forthcoming) | assigning blame inaptly |
| False Narratives (forthcoming) | bad analytics, and the bad storytelling after |