Hugging Face has achieved the unthinkable—they've amassed a collection of one million machine-learning models. This number likely exceeds the number of Starbucks out there or the number of unread emails in your inbox. In fact, that’s almost enough models to gift one to every person in Iceland, with one more as a bonus.

We. must address this with the decorum it deserves: fine-tuning and versioning, the gentle art of tweaking, have swelled the ranks of these models to a figure that might even make a mathematician blink. In other words, a significant portion of these models is merely minor tweaks to existing ones – i.e., less original than a Hollywood remake. The number dwindles to a far less impressive one if we're talking about truly groundbreaking LLMs. But, still in hundreds or thousands, and large enough that only machines will have any hope of going through all of them to assess their unique parts.

When it comes to counting, the Nobel Prize committee appears to have developed a new fascination for AI—almost like they’re handing out medals at a middle school science fair. DeepMind just bagged one for teaching computers to fold proteins better than your grandma can fold laundry—except it was awarded in Chemistry, not Biology.

Let’s not forget the physics prize that went to two machine-learning pioneers from the 1980s. Sure, their early equations laid the foundation for much of today’s AI work, but seeing it recognized as a Physics achievement? It’s as if the committee is saying, “Why not? If the equation fits, let's just call it science!” AI is supposed to be interdisciplinary in many ways, and with these prizes, one can raise that count by one.

Let’s go back to those models for a second and the best-in-class awards that help fill the tech journals these days more than the hardware that have refused to change for years. Trying to keep track of the best ones is like trying to follow the plot of a Christopher Nolan movie while simultaneously solving a Rubik's cube. The rankings change more frequently than a teenager's mood.

Which model is the best is one question where this analyst should be allowed to have an answer on any day that has no correlation to anything he may have said before. This leaves my long-suffering colleague, Venky, in a constant state of bewilderment. One week, I'm singing the praises of Model X; the next, I'm extolling the virtues of Model Y. Venky, bless his soul, merely sighs and reminds me that we are not, under any circumstances, to invest in anything for their AI-making capabilities as of today. Nothing in this field stays the same, even for hours.

Still, there is no dearth of private investors throwing money at LLM makers like it's confetti at a parade. The number of people who feel that creating AI is complex rivals the number of skeptics who have made up their minds about the uselessness of AI based on some history book reading and fiddling with ChatGPT in the most basic way a while ago. One wonders if these naysayers are overwhelmed by the sheer volume of options, much like a gentleman with an excessively long wine list. Perhaps they'd be more receptive if each model had a complimentary cheese pairing. (There's an idea for Hugging Face.)

Let’s turn serious before we end our millennial rant: Amidst this sea of models, some curious trends emerge. LLAMA, the undisputed king of the Hugging Face jungle, boasts a staggering 70,000 derivative models. Mistral, a relative newcomer, trails behind with a respectable 25,000. But the dark horse in this race is Alibaba's QWEN, quietly amassing 50,000 derivatives like a diligent squirrel hoarding nuts for the winter. One can't help but be impressed by its stealthy success.

In this era of instant copiability and rapid iteration, where a million models bloom and wither like flowers in a time-lapse video, one might be forgiven for feeling a sense of vertigo. But amidst this whirlwind of innovation, the core themes of our philosophy remain steadfast: Super Moore Growth, Commoditization of Ideation, Hardware/Software Flip, The Machine Era, and the dawn of the Fourth Macro Sector. These forces, like the steady currents beneath a choppy sea, guide us through the turbulent waters of the AI revolution. And as we navigate this uncharted territory with little visibility, we must remember that actual value lies not in the sheer quantity of models but in their ability to unlock human and machine potential and drive meaningful progress like those protein folding that came without any "killer apps".