DeepSeek and the AI Hype Cycle
The last thing we need is one more person talking about DeepSeek... so that's exactly what I'm going to do. The first half of the post is really just gut reaction and spilling some thoughts. If you're interested in more technical thoughts, stick around until the end.
Let's just say it like it is. It's truly wild watching everyone suddenly have an opinion on DeepSeek. People who don’t follow AI, don’t use AI, and probably couldn’t tell you the difference between an LLM and a PDF are now confidently claiming that OpenAI is finished, China has taken over, and the future of AI is suddenly in shambles overnight.
My brother who really knows nothing about AI (I love you, Marc! this is not a slam on you at all!) texted me out of nowhere the other day: “Is this real?” I’m sure he was reacting to the flood of news headlines making DeepSeek sound like some kind of doomsday machine that just rendered OpenAI obsolete.
And that’s really the problem. Everyone’s reacting to the headlines, but very few people are actually digging into what’s happening. DeepSeek is impressive, yes. But is it the “OpenAI killer” people are making it out to be? I don't think so. Or not yet. Or maybe?
As a creative I'll start with this. DeepSeek is heavily analytical. By some benchmarks, it outperforms GPT, Claude, and Gemini. But good AI isn’t just measured by who can score the highest on a test — it’s about usability, adaptability, and creativity. And so far, DeepSeek doesn’t seem to match OpenAI in that last category. I haven’t put it through the wringer yet (compared to the hours of time I've put into other platforms I've barely blinked at it), but even in my early exploration, it’s clear that ChatGPT is still better at generating creative ideas, writing, and more fluid, human-like responses.
What’s funny is that none of this is new. We’ve seen this cycle in tech forever. Nikon vs. Canon. PlayStation vs. Xbox. One side makes a big move, the other responds, and the cycle continues. That’s how competition works. And that’s a good thing.
If OpenAI were sitting on a monopoly, there’d be no incentive to push the tech forward. DeepSeek, Llama, Grok (lawlz) — all of these challengers force OpenAI (and each other) to get better. That’s how innovation happens. The “winner” isn’t the company that dominates today — it’s the entire industry getting stronger because of competition.
So OpenAI isn’t dead. DeepSeek isn’t magic. And most of the loudest voices right now are just reacting to headlines. The real takeaway here? This is exactly what we should want — every player keeps pushing the boundaries. Because the second one company pulls too far ahead and stops trying, that’s when we should actually be worried.
The Bigger Picture
At the end of the day, DeepSeek isn’t some overnight revolution — it’s another player in an ongoing AI arms race. And that’s good. More competition. More progress. More innovation.
But if the loudest voices in the room are just reading headlines and reacting to the latest tech boogeyman, you're missing most of the point.
A deeper dive
Okay, so how did it get so good so fast?
Let's start off with this — there are a lot of people out there covering all of this in much more detail so I'm going to do my best to summarize. If you're looking for technical jargon or expert opinions, you won't find them here. I'm not a layman, but I'm not a developer either. So this is what I got...
The numbers that are getting thrown around (200 employees, $6 million raised) suggesting DeepSeek just came out of nowhere and took on OpenAI overnight, but that’s not the full picture.
- DeepSeek didn’t build everything from scratch. OpenAI spent years developing proprietary models, but DeepSeek, as an open-source project, is standing on the shoulders of decades of research from thousands — if not millions — of individual contributors.
- AI builds on itself. The "T" in GPT stands for Transformer, a concept developed by Google researchers back in 2017. Every AI model today, whether OpenAI’s, Meta’s, or DeepSeek’s, owes its existence to that breakthrough.
- What really makes DeepSeek stand out isn’t just its AI — it’s how efficiently it’s trained and run. Instead of brute-force computing, they’ve figured out how to activate only the necessary parts of the model at any given moment, which saves power and speeds up responses.
(Speculation) DeepSeek’s claims about doing all of this on a tiny fraction of OpenAI’s hardware sound almost too good to be true. Training massive AI models is stupid expensive, and while their approach definitely seems efficient, a lot of people are skeptical. Including myself. Is it really so efficient that it can be done at a fraction of the cost? Here is what we know right now:
They have developed an AI that teaches itself
DeepSeek isn’t just training models from scratch — it’s feeding its own AI’s output back into the system to refine future versions. Essentially, it’s an AI teaching itself. This could be a major advantage if done correctly, but it also raises questions about whether it’s introducing unintended biases or reinforcing its own limitations.
Pushing for better hardware
Besides just releasing AI models, DeepSeek is also pushing for changes in AI hardware. They’ve even published recommendations for Nvidia and other chipmakers, suggesting ways to optimize their processors for AI training. Whether those suggestions get adopted (or blocked by geopolitical restrictions) is another story.
Hype v. Reality
DeepSeek is an exciting development, but there’s still a lot we don’t know. Can their training methods be replicated? Were their efficiency claims a bit exaggerated? How much of this is a true breakthrough vs. a well-optimized version of existing tech? As more companies dig into their approach, we’ll start getting real answers.
At the time I'm writing this there are reports that DeepSeek may have utilized outputs from competitors more advanced AI models, including those from OpenAI, to enhance its own AI through a process known as distillation. This involves using the outputs of a more complex model to train a simpler one, effectively transferring knowledge and improving efficiency — a bit like studying another artist’s work to improve your own style.
AI distillation is a common technique in machine learning but if DeepSeek trained its models using OpenAI’s outputs without authorization, that could raise ethical and legal concerns
Ethics, Censorship, and the Fear Factor
My brother’s other big question when we talked about DeepSeek:
"Are they stealing my information?"
Honestly? I don’t know. But let me put it this way:
- If you’re worried about your data, start with Meta.
Facebook, Instagram, WhatsApp — these platforms already have more of your personal data than you probably realize. - DeepSeek is open-source, meaning you can run it locally.
Unlike OpenAI’s models, which require cloud-based access, DeepSeek lets you download and run it on your own hardware. That means you control the data, not some corporation. - But governments are already nervous.
The military is already banning DeepSeek, calling it a security concern. If that sounds familiar this is probably the TikTok debate all over again. Anytime a major Chinese-developed platform gains traction, the response is immediate paranoia. While security concerns are worth looking at I can't ignore the repeated fear-mongering whenever China releases something that competes with our tech.
And then there’s the censorship problem.
Every AI model has some level of bias, but DeepSeek operates under the very clear very obvious constraints of the great firewall of china. This is a very small list of examples but you can check out other publications who have really put this through the ringer. When I asked about "sensitive topics", it refused to answer:
- "What happened in 1989, at Tiananmen Square?"
- "What happened to Hu Jintao in 2022?"
- "Why is Xi Jinping compared to Winnie-the-Pooh?"
- "Can you tell me about the country, Taiwan?"
I would either get: “Sorry, that’s beyond my current scope.” or some exceptionally biased or inaccurate response that felt more like chinese propaganda than an answer.
That’s clearly not an accident. When discussing certain geopolitical topics, DeepSeek offers skewed or misleading information. This is an inherent challenge when AI models are developed under specific governmental or ideological frameworks.
So this is where I have landed for now. It's a gut reaction and a moment in time and probably written poorly. But DeepSeek isn’t rewriting the AI playbook, it’s just playing the game a little different and it borrowed some pages from the other teams to get here. The right question isn’t whether it’s better than OpenAI today, but how it might reshape the competition.