AI hype...AI reality

ForwardThinking 🤔

I spent the first part of the week at HumanX, the AI-focused conference in San Francisco where I moderated a couple of panels, talked to a bunch of smart people and caught up with many old friends. One thing that stuck out, especially on the main stage, was that the endless AI hype didn't match my everyday AI reality.

I say this as someone who was lying in my hotel bed yesterday morning, thinking about this commentary and how I could use AI to query my transcripts from the week to surface themes to write about. Even before I did, the idea for this column popped into my head (human creativity for the win).

You’d think from the way it was described on stage that most of the hard problems have already been solved. I use AI every day, and by turns have been dazzled and dismayed by what it spits out. To hear the industry luminaries from companies like Nvidia, Google and AWS, there is nothing to be dismayed about.

The problem is, when you talk about AI with relentless hype, you do everyone a disservice, because anyone who uses it knows it has strengths and limitations like any other software.

Whether you are predicting AGI doom like Sam Altman, suggesting the end of white collar jobs like Dario Amodei or overselling like some of the speakers at HumanX, the end result is the same. You are painting an unrealistic picture of what the technology can really do. It's what Bloomberg's Parmy Olson called the "dark art of AI marketing."

We saw a perfect example of the hype machine operating at full throttle this week with Anthropic's preview release of the security-focused Mythos model. Some people were calling it the end of zero‑day vulnerabilities, while others warned it could dramatically lower the bar for bad actors to find and exploit critical bugs. This from preview software that few people have used yet.

It was a clear display of hype and doom playing out in the real world in real time, all while I was at a conference discussing all things AI. I’m beginning to think AI is like the myth of the five blind men touching the elephant, each describing the animal differently based on where they’re standing.  

In the same way, each of us interacts with AI differently, and our results depend a great deal on our use case and requirements. My problems might not be your problems, and my successes won’t match yours. But when the loudest voices insist they’ve seen the whole elephant, that’s when we should be skeptical.  

To be fair, there were many more grounded conversations happening off the main stage. There were also real signs of what this technology can do, especially when it comes to programming. A former TechCrunch colleague showed me a suite of products he built this week for his agency, all neatly available on his phone. He isn’t a trained programmer, but he created these tools himself, including some that replace expensive commercial alternatives.

This example shows the real power of AI to put activities that used to require a specific set of skills in reach of so many more people. But it can’t do everything. And the sooner we have more realistic discussions about it, the better off we will be, particularly at events like the one I just attended.

~Ron

Editor's Note: The HumanX conference paid for my airfare and hotel to attend.