Artificially Inflated
Read to the end to enjoy OpenAI getting mad at people stealing their work
Since the GenAI hype cycle arrived, I’ve been unsettled by the god complex on display by the current crop of overlords pushing the technology.
It’s an overwhelming, insufferable and mostly unjustified belief that they are heralding the new digital age, one that will transform every aspect of our lives by automating many of the jobs we do so we can live lives free of mundane work, supported by an economy that makes money for us on autopilot. Or by making us unemployed, removing our agency, killing our creative endeavors, and one day blowing us all up. (I’m a glass-half-empty kinda guy, so you know where I stand.)
It’s artificial inflation in action. A few prominent cheerleaders of the technology telling us over and over again, in tweets, blog posts, talks and podcast interviews, about how significant their work is, how necessary it is, and – this bit is important – how expensive it will be to build. We’ve been told that the production of these plagiarism machines is in the hundreds of millions, and the server farms that power them will run up into the hundreds of billions. Sam Altman is on record talking about the trillions of dollars his company will need to get the world ready for AGI. It is ironic coming from the guy who championed the company’s non-profit “AI for everyone” philosophy, only until it stopped him and his company from getting cash-rich from the VC money that was desperate to flood into the field.
It was all a bit… suss. I’ve always assumed a lot of people who were building AI products were grifters. That much was obvious when their social bios once had laser eyes, then mentions of the blockchain, then pixelated monkey pictures before finally landing on AI.
Did anyone ever stop to think these guys were full of shit?
And did anyone ever consider the grift went right to the top of the industry?
It seems not. OpenAI’s valuation is somewhere around $157 billion, despite losing money on every single query its chatbot answers and despite having no viable way to reverse this. They keep telling us the answer is that they need to keep growing, that they need more money, that they need more data sets, that they need more backing from governments, and that they need infrastructure built for them. And it seems the world was going to keep giving them it — see the recent Stargate Project announcement — because, well, they’ve told us so many times that it’s clouded our better judgment. We’ve been bent over and primed to receive, desperate to keep our AI overlords happy because they are going to dominate the future.
But this week, the generated script got flipped.
DeepSeek, a Chinese company, released a chatbot that is on par with ChatGPT, and it was (seemingly) built for a few million dollars — about 1/50th the cost of OpenAI’s model. More so, they managed to build it without the splashy fundraising, without the speeches at tech events, without constantly posting about how “transformative” it is and without going on podcasts to exercise their god complexes.
It just appeared, and all hell broke loose. I don’t know much about it, nor will I use it. There’s every chance that DeepSeek is doing what Western companies do; tweaking numbers and metrics and benchmarks to suit its messaging, and that it’s not as good as it says it is, or it wasn’t created for as little as they claim. The $5 million figure should be disputed, as it likely doesn’t contain salaries, failures and other associated costs. But the sheer size of splash the launch made (it immediately became the number 1 app on the App Store, ahead of ChatGPT) shows that companies like OpenAI are not as far ahead as they claim (or again, as they artificially inflate).
Whether it’s any good or not is beside the point here. It’s shown what others outside the Silicon Valley bubble are capable of, and it’s causing a stir. What’s more interesting is the ripple effects this may cause. To borrow from
–“Essentially, as long as DeepSeek’s fundamentals are born out, the research paper the company published holds up to scrutiny, and the app stays popular (and is not say banned by the state like certain other high profile Chinese-made apps), then what we are seeing can be considered a Great Undermining of the American AI industry’s foundational assumptions. Yes, in all bold.”
That assumption, of course, is that building AI has to be expensive, that the models need to get bigger to get better, and that it must be powered by data centers controlled by monopolies. That was the way AI was going to get built. Right? Whether DeepSeek is successful or not, or even widely adopted, it spells danger for the artificially inflated AI bubble. OpenAI is already charging $200 a month for some of its plans, and it’s still losing billions a year. When it’s struggling to generate revenue, and a competitor just dropped a model that’s free and of comparable quality, why would anyone pay for ChatGPT? Adoption of DeepSeek or similar products would prove that the average user is willing to spend $0 to use a form of AI. So, in that context, how does OpenAI plan to make money off this in the next few years? If competitors continue to improve, and industry-leading, free, actually open-source models flood the market, it would kill the company’s already shaky business model.
The markets panicked too, perhaps waking up to the reality that the tech stocks are propped up on AI Kool-Aid, and more specifically, an overly-inflated computer chip company run by a guy who signs tits like he’s some rockstar. In the aftermath of the launch of DeepSeek, Nvidia dropped 17% of its value, or $568 billion, which is a new record loss in market history. It’ll recover — and there are doubts as to whether the launch caused the drop; after all, the product was trained using Nvidia chips — but it reinforces how interlinked these few companies are with everything driving market momentum. If other countries do make products that don’t rely on these chips or take the logical step to create their own chips, what does that mean for the chipmaker and, worse, the entire stock market? It’s not healthy, and the reset can’t be far away.
Let’s end on a more fun note. OpenAI is clearly rattled, and its response to the launch of DeepSeek is the most glorious dose of irony you could ever hope to serve somebody: The plagiarisers are mad their work got plagiarised!
This one from 404 Media made my day —
And this one, which I saw on
’s latest post —The best thing here is that the company is being blatantly hypocritical and likely undermining the very arguments it’s making in defense of itself in a copyright lawsuit with the New York Times. It’s shining a light on its own shitty, entitled practice of stealing the world’s work without asking, saying that intellectual property matters and that you can’t just steal someone else’s work.
Finally, something we agree on.
As someone in the UK, whose government is considering *changing copyright laws* to allow LLMs to take our IP unless we opt out, the schadenfreude over this is just glorious!
Haha. DeepSeek didn't just steal OpenAI's IP. It stole every frontier model's algorithms to arrive at a hybrid model using best of each. Probably did do it for $5m. All courtesy of the Chinese governments best and most talented industrial espionage tools and capabilities. Nothing new to report here...use this free AI model that is better and faster than all these other pay as you go LLMs so we can get our meathooks into your personal data and IP to go along with what we just stole from your (our) venture backed, tech titan masters.