180 - Performance vs. Innovation

Watching the latest LLMs come out is a lot like watching RuPaul's Drag Race. You know that their objective is to look the part, and however well or poorly they perform, nothing under the hood really changes.

Of course, the difference is that RuPaul's is for entertainment value and personal competition. Nobody is attempting to throw hundreds of billions of dollars at making ever larger drag shows, or 1,000x scale Drag Queen Colossi.

Oddly enough, the AI domain has developed another eccentricity over the past few years that closely parallels that of a Drag Show, in that the only ones competing on the runway are explicitly not what they aim to appear as. Just as you explicitly won't find a pair of X chromosomes walking down that runway, you won't find anything cutting-edge coming from LLMs.

Like a late-stage Alzheimer's patient, many in AI now focus on scaling up systems, only to rapidly forget how badly those systems fail to live up to expectations at each new scale. More than a year ago a number of cons were claiming that combining Wolfram Alpha with an LLM would magically create "AGI". Of course, nothing of the sort happened, because LLMs are almost exclusively based on hype, as they bring nothing new to the table.

Nvidia likes to claim that they sell shovels in a gold rush, but a more accurate comparison would be that they sell munitions in a genocide. Most of the companies reaping the greatest profits and raising the most funds today are willful accomplices in some of the worst crimes that humanity has ever seen, and may ever see.

Fortunately, I can say this about them because their hardware has never been a bottleneck for technology at the cutting edge, where brute-force compute isn't the driving factor. They hold no say in the future of technology, as no amount of posturing changes the fact that they're empty-handed when it comes to the most advanced AI and the hardware required for it.

The next time an executive talks about integrating an LLM for your company, take them to the nearest Drag Show. If why you've taken them doesn't dawn on them before the end of the show, quit or get them fired, because a company led by that level of stupidity has no future. Leave drag to the RuPauls of the world, rather than integrating it into your business-critical systems.