Author: IRPA AI Senior Analyst & Former WH Transformation Advisor, Chris Surdak
If you have been hiding under a rock over the last two weeks you may have missed the latest craze on Linkedin: using generative AI to make fake action figures. I have seen several hundred of these images all of a sudden, and most of them are far from complimentary. What does this latest craze in the use or abuse of LLMs reveal about the future of this technology, and more importantly, what does it say about us as their users?
These memes, displaying a toy “action figure” following the guidance of the user, is the latest in the explosion of users leveraging the capabilities of generative AI to produce, well, noise. It’s attention-grabbing, and occasionally even entertaining, but is this really what we’ve invested a trillion dollars to achieve? And if these tool are capable of more than mere entertainment and sarcasm, where are the compelling use cases to this end? I hear many people claiming that AI is “transforming” how they do business, yet they can rarely if ever point out specific ways in which this is occurring or specific end results which represent a better, cheaper or faster alternative to whatever means were available before generative AI.
Similarly, there’s a great deal of hype lately about “vibe coding,” where people are claiming that they create entire businesses, in minutes, simply by having a chat with one of these platforms. The LLM cranks out mountains of code meeting the descriptive needs of the user, and it all magically works without any testing or validation. Excuse my doubt, but this sounds not only implausible, but also extremely dangerous.
Software doesn’t work this way. Indeed, most software doesn’t work until about two or three hundred iterations of trouble shooting. Further, as someone who front-ended the development effort of a few hundred software deployments in my career, the notion that a customer could effectively articulate what they wanted, sufficient for an AI to go and build it, seems similarly implausible. The vast majority of actual software projects usually end with a user saying, “You built what I asked for, but not what I needed.” No amount of clever applied statistics (AKA, LLMs) is going to correct for human errors in requirements definition.
Collectively, meme generators, vibe coding and AI Agents arguing with each other on the internet, are all going to contribute to the Grey Goo problem. More and more training data is required to feed these beasts, and more and more of the available content is being generated by these beasts themselves. If LLMs are Garbage In Garbage Out (GIGO) factories, and they are, then the more garbage they produce, the worse the situation is likely to become. I’m not certain that there is a solution to this issue, however we must wonder if such data pollution may grow so bad as to make these tools unusable, rather than just wrong half of the time.
About the Author:
Chris Surdak is a Senior IRPA AI Advisor and was formerly White House Chief Transformation officer, Automation & AI Practice Lead at EY & Executive Partner for Digital Transformation at Gartner. He’s an engineer, futurist, transformation executive and best-selling author, with over 30 years’ experience in technology development and deployment, digital transformation, blockchain, data and analytics and AI & intelligent automation.
CLICK HERE TO SCHEDULE AN ANALYST CHAT
Links:
Originally posted in the IRPA AI Network — Announcements & Updates