Three Years In: Who's Winning with LLMs

This was a weird post for me to author. The idea started as a sarcastic piece because that’s my coping mechanism. But partway through the ideation process, I realized that I genuinely feel this ranking is accurate. So, let’s have at it.

It’s been about three years since large language models (ChatGPT, Claude, Gemini, etc.) went mainstream. We’ve seen billions invested, trillions in market capitalization created, and, depending on who you listen to, the future of human civilization is at stake.

But if we want to be truly honest about the value of a piece of technology, we need to take a broad view of its impact. As an example, the nuclear bomb is a horrible creation, but nuclear power can provide scalable, relatively clean energy that could, in theory, replace the majority of fossil fuel use.

That said, here is my ranking of who is currently benefiting the most from the proliferation of these tools.

1. Nvidia

Nvidia sells shovels. Every gold rush needs a shovel company, and Jensen Huang showed up in his cool-tech-bro leather jacket and is charging 5 figures per unit. The best part for them is that, like the gold rush in the 1800s, it doesn’t matter if any of their customers actually find gold. The shovels are sold, and the revenue is booked. Something I see overlooked on social media is that Nvidia doesn’t actually own any fabs, so a bubble pop is unlikely to do any lasting harm to them.

2. Venture Capitalists Who Needed a New Cycle

I’ve spent much of my career in the startup space, and I’ve worked with some quality VCs and some… not-so-great ones. The blockchain and crypto scammers jumped right on the AI hype train. The playbook is to raise other people’s money, take their fees, fund the hype, extract at the Series C, and hope to dupe pension funds and institutional investors into holding the bag.

On a side note, OpenAI is seemingly struggling to raise funds. My cynical prediction is that, because their failure would trigger a fast and massive correction, they will be extended a lifeline by big tech to get them to IPO. Then, hopefully enough rubes buy in to hold the bag, and hope that the bubble deflates slowly enough to wipe out retail investors over time rather than bursting outright.

I’m honestly undecided as to whether the coming correction will be a scream or a whimper. There will be a correction, but timing the market is for fools and gamblers.

3. Sexual Deviants

The European Commission estimates that around 98% of deepfakes are pornographic in nature. Think about that: the single most prolific use of the most powerful image and video generation is used to make nude images of people who didn’t consent to it.

This is not the “consumer-led adoption” that people are looking for.

And before some AI-hypester comes in and says, “Well, you can make deepfake porn with Photoshop”. Yes, you can. You can also hand-draw porn. I’m honestly exhausted by debating people who don’t think that ease of use and scale are differentiating factors.

4. Spammers, Scammers, and the Entire Bad-Faith Communication Industry

Writing a convincing phishing email used to require at least a passing familiarity with English. That barrier is gone. Spray-and-pray is now spray-and-pray at volume, in any language, with perfect grammar and a personalized opening line referencing your dog's name from Instagram.

As a career technologist, I can assure you that, as a society, we are unprepared for the massive wave of fraud coming, especially given that “Agentic AI” cannot solve the prompt injection issue.

A problem that has plagued software since the beginning is that data and instructions are mixed. This is why remote code execution is a thing that we try to guard against. LLM agents not only mix data and instructions, but they often pull data from untrusted, external sources. If you’re in the know, this is a terrifying development.

5. The Surveillance State

LLMs are genuinely useful for summarization. However, this also means that every authoritarian government just received a force multiplier for processing the firehose of data they were already collecting.

Summarizing large volumes of intercepted messages used to require teams; now it requires only a prompt and a contract with Palantir. Sleep well.

I can’t even express how insane it is that they had the balls to name their company Palantir.

6. Propagandists

Astroturfing used to be labor-intensive. You needed troll farms, and troll farms needed rent and bathroom breaks. Now, one person with a laptop can simulate a grassroots movement before lunch and a counter-narrative after dinner. It’s genuinely difficult to figure out what’s real and what isn’t.

I don’t know how many more “Russian Bot Farms Exposed” pieces I can read before I just start recommending people get off social media.

7. “Those” Consultants

You know who I’m talking about. LinkedIn is loaded with people who sell confidence and hype over competence and measurable outcomes. They now have a shiny new toy to sell, and big tech is running cover for them with massive amounts of FOMO.

“You’re falling behind!”

“You won’t be replaced by AI, you’ll be replaced by someone who uses AI.”

“I made X in Y hours; people in the Z field are cooked.”

I’ve been asking for receipts for 2 years now, and no one has delivered. This is probably the most infuriating aspect of the LLM hype in the business world. Productivity isn’t massively up, the number of software releases isn’t substantially up, bugs and tech debt are noticeably compounding, and none of the big providers can point to a blueprint of how to achieve ROI with their tools, despite having access to every prompt put into them.

I’ve had people ask me what it will take to get me on board with the new wave of AI tools. Evidence. The answer is always evidence.

8. Business “Leaders” from the Jack Welch School of Management

I hope that younger generations of students and workers are paying attention to just how… gleeful some members of the executive class are at the thought of no longer having to pay those pesky wages.

First, they march on stage and announce the layoffs, then they quietly start hiring humans back after their services and resulting customer satisfaction tank. Salesforce, Klarna, and Duolingo are all notable examples of this trend.

For a while, investors were rewarding layoffs due to AI. Today, that trend is reversing as executives try to blame AI for natural economic cycles and poor management. I want every business journalist, when faced with an executive claiming layoffs are “due to AI,” to start asking for specific examples of how AI has delivered.

And no, Microsoft, Amazon, and Oracle, laying off workers to fund CapEx for your soon-to-fail data center investments, don’t count.

9. The Posers

There are many people who are either too unmotivated or too untalented to create value. These people are very, very excited about AI tools because they’re convinced that effort isn't necessary for value creation.

• People who can’t write think they can produce passable copy.
• People who can’t code think they can produce professional applications.
• People who can’t think can pretend to have compelling arguments.
• People who can’t think that using a plagiarism tool with token karaoke is worthy of praise.

The fun thing about the internet and algorithms is that some of these people will have some success. There are already grifters taking advantage of this, just like other fads: Pay for my course/blueprint on how to make massive money spraying slop all over the internet.

10. Lazy Students

The posers are all grown up. They start as lazy students. They don’t want to learn or compete. And look, I’m going to soften up on these kids because their brains aren’t fully formed yet.

Learning, especially rigorous learning, is uncomfortable. Children tend to avoid discomfort. Well, so do adults, but kids get a pass. The adults in the room are supposed to be the ones encouraging discomfort because they understand the long view, whereas most kids would rather play Call of Duty than learn linear algebra or write essays.

I’ll get to teachers in a moment, but the point here is that these tools make work avoidance so much easier than ever in our history. I feel really bad for the current generation of learners, and I’m genuinely concerned that this is how we end up with Idiocracy.

11. Lazy Teachers

Again, this is where we need some nuance, but the state of education, especially in the United States, is in decline. LLMs have exposed just how much performative bullshit is present in what we refer to as education.

Grading is tedious, and feedback is hard. Genuinely engaging learners is a skill. Teachers are dealing with reduced budgets, larger class sizes, and now AI grifters are showing up to claim their tools are the solution.

The more likely outcome is that AI graders will grade AI papers, and few will learn real skills. This will eventually ripple through the job market and the economy. These types of generational issues take a while to manifest and are damn near impossible to fix when they do.

AI is going to finish what social media and Chegg started.

If you’re keeping score, you’ll notice that the people who were supposed to benefit, workers and businesses trying to do their jobs better, didn’t make the list. They’re on the list, just way further down.

As always, I’m not suggesting that LLMs have no utility; in the right hands with the right focus, they do. But here we are, three years in, and I’m seeing way more harm than benefit. Consider that I didn’t even get into the ethical issues surrounding how these models were trained on copyrighted materials, the legal profession's growing concern about hallucinations in real court cases, and the environmental impacts of these data centers.

But hey, at least we can generate pics of Elon Musk in a bikini, right, Grok?