Lenin never read this Homer
“There are decades where nothing happens; and there are weeks where decades happen.” — Vladimir Ilyich Lenin. This is one of those weeks. “Two wrongs [do] make a right.” — Homer Simpson.
OpenAI’s Sam Altman claims he has evidence that DeepSeek used its model for training.
As versions of that headline ricocheted across the internet, phrases like unmitigated gall and blinding audacity came to mind but nothing quite summed up Altman’s dishonest, disingenuous duplicitousness quite like Homer Simpson did in season 4, episode 22, “Krusty gets Kancelled.”
Krusty loses his show to a ventriloquist and a mannequin named Gabbo. Gabbo crank calls the now-out-of-work Krusty. When Krusty figures out the gag, he yells into the phone, “If this is anyone but your Steve Allen, you’re stealing my bit.” Lisa and Bart squabble over who’s right. Homer settles it by telling them, “Two wrongs [do] make a right.” What can I tell you, I’m a simple man. I read The Iliad in comic book form.
Let the media prosecute the players in the court of public finger-pointedness. It doesn’t matter. They’re both liars and thieves. Besides, there’s a bigger game afoot. The business of AI just got Netscaped. And, in this case, that’s a good thing.
The business of AI just got Netscaped.
Netscape started in 1994. They built a browser that let us travel the world wide web and they sold it us. Yes, in 1994, we still paid for software. Netscape went from being worth zero to being worth three billion dollars in just sixteen months. Their revenue doubled every quarter. Because the internet was the razor and Netscape’s browser was the blade. A week later, Microsoft made the Internet Explorer browser and gave it away for free. Netscape market share went from 90% to less than 1%. In that week, a decade happened.
DeepSeek’s origins don’t matter. The US Navy telling their people not to use it due to, “security and ethical concerns” don’t matter. It doesn’t matter if DeepSeek can answer a question about Tiananmen Square. It can’t. It doesn’t matter if DeepSeek is better or worse than other AIs. It doesn’t even matter if DeepSeek is a CCP data gathering perpetual motion machine. It probably is. The only thing that matters is that DeepSeek is free and open; and OpenAI is paid-for and closed. DeepSeek is Internet Explorer. Every other AI is Netscape.
To be fair, I didn’t notice the problem until I thought about the tweets from Perplexity’s CEO, Aravind Srinivas. For the past twenty hours Aravind’s has been as out of step with the AI community as Paul was from John, George, and Ringo on the cover of Abbey Road. It’s clear now. Perplexity hosts a version of DeepSeek. Think of it as Duck Duck Go for AI. Duck Duck Go may seem like its own search engine. Really, it relies on Microsoft’s Bing for many of its answers. If you use Bing, Microsoft knows who you are. If you use DDG, they don’t tell Bing who you are. You’re masked. That’s what Perplexity is offering. A masking service for DeepSeek. That may be fine for people. But it doesn’t go far enough for businesses. So, the problem still isn’t solved.
About a year ago, I was chatting with a senior executive of Chinese based pharmaceutical company and I asked if they could use AI to expedite drug discovery or development. “Nope. Our research is far too propriety to share with other companies. We can’t risk the AI mixing our data with theirs.” Simply put, you lose agency over your data when you use any AI. Your questions and other data go into a black-box so complicated even it doesn’t know where your data ends and someone else’s data starts. Now add the Chinese angle and the heads of IT’s heads begin to explode.
In theory, companies can run their own version of DeepSeek’s inference engine on your own machine. None of the data goes back to DeepSeek’s servers in China. So, again, in theory, corporate data is safe. Safer even than if they had used OpenAI. Or a Perplexity mask. Because… it’s on-prem. “On-prem” is the oft-uttered catchphrase of my three most techy friends. In case you haven’t guessed, the prem is for premises. Your code running on your services in your building. But this isn’t just code. It’s open-sourced code.
Somewhere in the back, an IT person who works for a big bank (Hey, Paul!) is muttering, “My bank isn’t going to use some open-source software that isn’t backed by a big company. No bank would. You don’t get fired for buying IBM. You can get fired for using free code.”
And, if this were 1990, when I interned for IBM, Paul would be right. Except there’s ample precedent for on-premming AI. These days even IBM boxes run open-source software. Chalk it up to these two words, Red Hat. If you use open-source Linux (and lots of banks do) it’s supported by consulting services and products from Red Hat. IBM paid $34B to buy Red Hat about five years ago.
Now that you have an open-source AI, you need an on-prem AI consulting service. The new company needs a cool name though. Auto Preminger. No? Prem Hat? Personally, I like Acai. Maybe short for Autonomously Contained AI.
Once a market has been Netscaped it can’t go back. No browser launched since Explorer came out can use a paid model. Their business had to rely on something else. So when you read DeepSeek might be faked – and you will – know that it doesn’t matter. If they want to know how you know, say “Green Avocado phones.” And tell them this story.
Remember when phone companies were a regulated monopoly and you paid a hefty monthly fee to rent an avocado green phone that hung on your kitchen wall? Then, deregulation. You could buy any cheap phone and Bell had to let you connect it to their network. Bell’s pricey phone rental business was toast. Bell got Netscaped before Netscaped was a verb. Or a noun.
Every AI will have to give their services away and charge for other services.
In the meantime, I’m just going to leave this here…
Six months ago, I made this infographic and I wrote a story that it’s not the size of your AI, it’s how you use it. The value was in the application layer — what I called Smart Services — not the AIs themselves. I don’t get many wins. I’m taking a victory lap.
So, while newsies write about who stole what from who, the real story is that two wrongs made a right. Sam Altman was wrong to train OpenAI on other people’s work. Chinese hedge fund billionaire, Liang Wenfeng, was wrong to train DeepSeek on OpenAI outputs. Together, they gave us a roadmap to commercialize AI. The AI will be free. You’ll pay for the services that surround it. Instead of AIs buying chips from Nvidia, companies and their premming services will. So feel free to start buying Nivdia stock and nuclear reactors again. Whew.
DeepSeek came out last week. It only feels like a decade.
