Goodbye ChatGPT, Hello DeepSeek
- Calvin Stevens
- Jan 27
- 6 min read
Within a mere week of its launch, the app for the Chinese artificial intelligence model, DeepSeek, overtook its competitors — namely OpenAI’s ChatGPT, the common household name of AI — to become the most downloaded app in the US.
But what, exactly, is DeepSeek? Where does it come from? And why is its sudden rise so significant?
Background
DeepSeek is a Chinese Artificial Intelligence (AI) company, based in Hangzhou, Zhejiang, that develops and produces opensource large language models (LLM) — the same technology which ChatGPT and most other AI chatbots are currently based upon. The company, itself, is funded by a sole Chinese hedge fund, High-flyer, who reportedly only invested roughly 5.7 million USD and 2 788 GPU hours to fully develop and train the V-3 model.
In comparison, ChatGPT’s latest model, GPT-4, has been revealed by OpenAI to have costed around a whopping 100 million USD, taking between 90 and 100 days to fully train.
R1-zero, DeepSeek’s latest model, was trained almost exclusively using reinforced learning. This essentially means that the AI takes in actions through a given environment, and once these actions are interpreted into a reward and state representation, they are then fed back to the AI model. Over time, the model thus learns to meet a certain goal by way of a reward-punishment system.
The DeepSeek app
Lately, users from around the world have be flocking to the new app with the latest DeepSeek model (including, and particularly, African tech-gurus).
The user interface is sleek and simple, containing all the beloved elements users experienced with ChatGPT, making it very user-friendly. Moreover, it comes with an extension to search the web, giving it the feel of any regular search engine.
When users download the app, they will be greeted with a friendly, familiar face.
But despite it’s relative simplicity — and keep in mind, the app is barely a week old in the public’s eye — there are a number of reasons, cited by users, for it’s superior feel to ChatGPT.
And all lie within the model’s functioning and capabilities.
Some users have expressed their awe for the model’s smooth web-searching abilities, criticizing OpenAI’s lack of similar such models.
Others have pointed to the model’s reinforced learning feature, calling it a “game changer” and “the future” of AI. A bonus, here, is that the model is able to concisely display it’s step-by-step process for obtaining the output from the given prompt.
The model is also quite the charmer — more so than ChatGPT — loaded with a sense of humour, a love for puns, and the ability to engage in a more human-like fashion (trust me on this one; a friend and I were testing DeepSeek earlier today, and needless to say it was able to hold an almost hilarious debate regarding who Bella should have chosen as her soulmate in Twilight. Spoiler alert: DeepSeek likes Jacob).
But there has been one universal consensus above all else: the cost.
Some concerns with the app
Despite the overwhelmingly positive reviews and feedback by users, there have been a few concerns. Most of these revolve around the censorship worked into the AI model.
The company’s founder and CEO, Liang Wenfeng, has been known for his connections with the Communist Party of China (CPC), and this has some users concerned with the what information can be retrieved from the app. For instance, when asked about the the Xinjiang Uyghurs, the model initially produces an answer but quickly deletes it and replies with the message: “Sorry, that’s beyond my current scope. Let’s talk about something else.”
Currently, the only way to remove the integrated censorship mechanisms is through an open-source version of the R1 model.
Otherwise, most prompts generally produce more detailed, and more comprehensive results than ChatGPT. Whether a split between a Chinese version and a “Western” version of the app will come — similar to TikTok and Douyin — remains to be seen.
So what’s the big deal?
Well, as one might conclude from the aforementioned paragraphs, the cost to develop the AI model, when compared to its more well-known competitors, is sending waves throughout not just the AI industry but the tech industry as a whole.
The seemingly low cost to develop an AI model on par with GPT-4 is, simply put, astounding.
Just last year, OpenAI’s CEO, Sam Altman, surmised the cost to uphold and boost AI infrastructure (namely through AI computing chips) would be upward of 7 trillion USD.
But the sudden disruption by DeepSeek has thrown investors into a panic. If all it costs to train an AI model (one on par with GPT-4, mind you) is 6 million USD, then why are US tech-giants struggling to accomplish the same feat with more than 10 times the money? A further 7 trillion dollar investment, then, seems entirely wasteful.
As a result, many tech companies saw their share prices within the stock market drastically fall over the course of this last week. US tech-giants, Meta, Google, and Nvidia, all saw their prices plunge. Even European companies with stakes in the sector and its peripheries succumbed to a similar fate. ASML, a leader in producing semiconductors, for example, observed a near 10% drop in their share prices, while a similar decline occurred for Siemens Energy.
The future of AI chatbots
Amidst the chaos that DeepSeek has stirred, a single question has arisen: is the US capable of maintaining its lead within the industry?
The way I see it, the current answer that stands is a big fat no. Not unless US companies can convince investors that spending all that money is truly worth it.
It does also bring said money into question. Where is it all going? How come China can accomplish this development with so little money and sanctions placed on their ability to buy US computing chips (Nvidia)?
Two general camps have taken root over the discourse.
The first argues that China, and particularly High-flyer, is purposefully underestimating the money invested into DeepSeek to disrupt the global market. Some AI experts claim that the 5.7 million dollar investment only comes at the cost of borrowing and repurposing already developed AI systems. The argument, essentially, is that China develops their models off of stolen data, likely US data.
The second camp, meanwhile, argues that the sanctions placed on China have actually been the catalyst for such a miraculous development. The sanctions have unintentionally caused a much larger collaboration between internal Chinese tech-companies, possibly resulting in a common resource-sharing platform. What’s more, DeepSeek, since its founding in 2023, has primarily taken on fresh graduates, curating a healthy young talent pool that is eager to prove itself. And currently the company is “solely focused on research” with “no detailed plans for commercialization”.
Which camp is correct?
Well, there is no definitive answer there, and the likelihood is that it’ll remain a two-way street for the foreseeable future.
I, myself, however am leaning more towards the latter camp; because even if — and I don’t doubt there is a chance of misinformation — that DeepSeek did not, in fact, cost nearly 6 million dollars to train, I cannot fathom what the point in drastically lying about the cost might bring. If I were to lie about my investments in an industry that has seemingly spent hundreds of millions, I would be a lot more subtle, more realistic, and “undercut” my competitors by a smaller margin. It would certainly seem more believable that way. But only 6 million? No, that number seems far too small for it to be a bluff.
Furthermore, I am of the opinion that a large portion of the arguments that arise from the first camp are clouded by both fear and a heavy anti-China sentiment. The chances of a relatively young company — who is invested, first and foremost in research — coming up with an innovative solution to producing a new AI model doesn’t seem all too far-fetched. The recent encroachment of large US corporations on the US government (notably with Trump’s inauguration) also has me skeptical of the allocation of AI funding — because, seriously, where do we expect 7 trillion USD to go?
So, needless to say, I’m not at all surprised at the sudden rise of a Chinese company within the AI industry. Whether the West likes it or not, China is massively up-scaling their innovation and technology; sooner or later, they will surpass even the US in certain sectors, if not already.
The best outcome, I hope, is that we see a greater push by Western countries to compete with the Chinese rather than dismiss them. Or, better yet, collaborate with them (although this, I fear, is less likely). Throwing around lies and remarks simply isn’t going to make the truth go away. The race is on, and China is taking the lead.
If only we get more healthy competition out of this, then perhaps there is room for even greater global innovation and development.
Commenti