It's really not clocking for people that this is the worst AI will ever be…

The below is my stream-of-consciousness thinking/writing after a recent LinkedIn post from me has gone somewhat viral with 20,000 impressions, and I've been somewhat surprised with some of the arguments regarding what AI can and cannot do today.

People really think that this is still just a chatbot, and I’m getting concerned.

I keep trying to convey to ordinary folks, physicians, and other professionals how those of us deeply involved in technology are using AI and the things we’re having to do. It really just doesn’t click for the average person how advanced these things have become. People really still think they’re just chatbots.

And it’s crazy, too, because the longer physicians and others take to hear what us technical people are telling them, the more likely the technology will have changed their field before they’re able to build any protections and insulations.

Why is it so hard to convey to the average person that you will simply not beat any of these AI models in knowledge and communication? It’s a losing game. Get over it. It’s done.

You’ve got people who barely want to go to the doctors but are addicted to talking to chatbots all day. And that was a technology from years ago, not the modern stuff. Wait until it starts actually using their computers and ordering stuff for them, and wait until we integrate them in video cameras.

I’m not fearmongering. I’m just telling you where things are headed and hoping that you start preparing in advance for it, and giving you ideas for how to do that. But people are so hung up on still thinking that AI is simply a chatbot because that’s what they used once in a while—because they don’t have the time to actually look into it deeply the way we have. Yet they keep doubting what we tell them because it’s too hard to believe.

Thiago Forte had a really interesting point a long time ago when he made his personal knowledge management system. He said, “The human brain was never meant for information retrieval. It really was not evolved to do this. It was meant for idea and information generation.”

We have stochastic randomness built into our senses, our five senses, that allow us to unpredictably filter information, ignore some things, and focus more than we should on others, and this leads to novel insights. That is the only fundamental advantage that humans will have as machines get better and better. It’s not even about AI; it’s simply about circuitry versus imperfect biology, and the advantage of imperfect biology is randomness. It’s like Darwinian evolution all over again.

The randomness that we have with how our senses perceive information is our main value-add. In Darwinian evolution, the randomness is in the context of mutations and survival of the fittest. In this case, it’s in the context of idea generation, information generation, data generation—the fundamental basis on which you eventually choose how to execute and act. That part, however, the machines are now better at and are increasingly getting even better at.

Maybe I need to refine my argument because at this point I feel like I need to yell from the rooftops so people finally hear and start acting differently and preparing for this in this way.

The biggest issue with this argument is that I’m realizing it’s not even necessarily true, because anybody who has any technical understanding of how these AI models work understands that stochasticity is actually also built into the LLMs as well. They have sampling temperatures, dropout, ensemble diversity, and different ways to build randomness into them.

Honestly, the more I think about it, quite literally, the main fundamental thing that persists is not even embodiment, because I do think with technology a lot of these machines will become embodied as well. I think it’ll come down to simply accountability and stakes. That is simply the only thing left.

That being said, there is an interesting counter-narrative happening right now, where people are noticing that a lot of the outputs from these LLMs are all very good, but are all hitting a plateau. Human ingenuity and creativity is no longer driving advancement and progress nearly as much as the foundation model companies themselves. We’re only advancing when they advance, and so we’re being pulled by them, which is causing a lot of production output to be somewhat the same.

If this pattern holds, people might start getting tired and bored of LLM outputs, and we might reverse course. But I suspect it won’t hold, and I suspect that we’ll get used to this new pattern. We’ll simply start having to dream an idea bigger, and outcomes will become unequally distributed once again.

Because if there’s one thing that’s been surprisingly revealed as part of these past few years, it’s that equal outcomes have actually been a pretty bad scenario. Unequal distribution of outcomes and opportunities seems to have been a much better producer of quality products than a higher quantity of equal output products.

That is to say, if everybody’s making the same AI slop, even though it’s good AI slop, it’s all the same one way or another. We have found out that this is actually a less superior outcome than having a few individual, brilliant—sure, you can argue privileged—creators building exceptional products that everybody else uses and follows along.

This is where we are. The technology has moved far beyond what most realize, and the real work now is preparing—not by denying its power, but by focusing on the narrow but critical ground where humans still hold the final line: accountability, stakes, and the courage to dream bigger amid the slop.