Grok Hilarious Fiasco Is A Serious Reminder

Grok is the Elon Musk version of AI, a chatbot that is supposed to be less woke. But lately it has also been a hilarious Elon Musk fanboy that will always tout the awesomeness of its owner.

It has boasted that Musk is “among the top 10 minds in history, rivaling polymaths like da Vinci or Newton.” Also, he would beat Mike Tyson in a fight. Bruce Lee, too. Given the choice between Musk, Peyton Manning and Ryan Leaf in an NFL draft, Grok said, “Elon Musk, without hesitation.” Musk is lean and muscular with extensive martial arts training, says Grok. Given the choice between switching off Musk’s brain and wiping out the entire nation of Slovakia, Grok would vaporize Slovakia.

And that’s before we get to ruder stuff, like Grok’s assertion that Musk could be the best in all of human history at drinking piss and performing blow jobs.

People started goading Grok with prompts like those asking Grok to praise Musk’s ideas that he didn’t actually have (e.g. historical theory about England’s break with the Catholic church). At this point it’s such a popular game that I’m not even sure if other examples are real, like Musk could be raised from the dead faster and more efficiently than Jesus-- but it sure seems they could be. Musk’s crew, for its part, has been scrubbing away the brown-nosing Grokspeak, and Musk himself posted on X that “Earlier today, Grok was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me. For the record, I am a fat [expletive].”

This is all a reasonably hilarious reminder of how over-the-top lying can get, as well as a look at how tragic it can be coming from someone who has more than enough resources to feel that he is--and has--enough.

But it is also a reminder that one of the oldest computer principles still applies to one of the newest advanced technologies--

GIGO. Garbage In, Garbage Out.

Decades ago we used to hear those words as reminder that you have to be careful about what you feed your program because otherwise it will give back junk. Nowadays, we have to confront another aspect of GIGO-- when the people in charge of the program feed it garbage on purpose.

It was enough of a signal when Musk announced that he would make Grok “less woke,” but now we’ve got a demonstration of how deliberately unmoored a chatbot can be, and not just because it’s in its nature to make shit up, but because it can be “adjusted” to make shit up to fit a particular bias.

I am daily frustrated by people who have fallen into the notion that AI chatbots are somehow paragons of objectivity that scan the web for the best evidence, weigh it logically, and deliver an unadulterated evaluation of the some total of human knowledge on whatever you have asked. They aren’t, and they don’t, and, in fact, you can “adjust” them so that they will always tell you that a certain man-baby is the most awesome human being to walk the earth--literally.

Understand that this particular Elon-centric “adjustment” is so over the top that almost nobody would mistake it for an objective True answer. But you must ask yourself-- how many more subtle and less obviously bonkers “adjustments” could be made to the program that would not be obvious to you at all?

Even if you are deliberately trying to create an unbiased program, you will fail, because your own biases are reflected in everything that you “understand” about the world and what is in it. The notion that you can build a machine out of your biased pieces of mental lumber and come up with a house that is perfectly square is silly (also, the idea that perfectly square is how to build a house is yet another bias).

On the other hand, if you would like to build an AI chatbot machine that was biased in your preferred direction--well, that is totally achievable. It might take some practice to build the bias in subtly enough that you don’t get caught, but with a little practice on top of huge amounts of wealth and compliant underlings, I’ll bet you can get there.

We’ve talked a lot about AI as plagiarism machines and bullshit extruders and cognitive automaters, but we should be sure to include high-powered lie generators on that same list. Because as long as humans are on one end of the machine accepting output as an objective representation of reality and powerful folks are on the other end scripting the objective reality they would prefer, we are looking at some toxic high-tech GIGO.