DeepSeek shook the markets. Is this the end of American dominance in AI?

 

"Nasdaq 100 futures fell sharply on Monday morning. An artificial intelligence model developed by Chinese company DeepSeek has threatened American dominance in the field of AI.

 

Chinese company DeepSeek released its V3 artificial intelligence model in December, and on January 20, it premiered an improved model called R1.

 

Nasdaq 100 futures fell by as much as 2.5 percent on Monday morning, while the S&P 500 fell by nearly 1 percent. It looked like Monday's session on Wall Street would start with a sell-off of American technology companies. The clear deterioration in sentiment on the stock market was explained by the debut of an artificial intelligence model developed by Chinese company DeepSeek.

 

Within days of its launch, the Chinese competitor ChatGPT jumped into the ranks of the most downloaded applications. Experts say DeepSeek can do more while using fewer resources, making it cheaper than the most popular AI tools.

 

What has DeepSeek done?

 

Chinese company DeepSeek released its V3 AI model in December, and on January 20, it premiered an improved model called R1. It quickly became one of the top ten best-rated AI models. "DeepSeek R1 is one of the most amazing and impressive breakthroughs I've ever seen," tweeted Marc Andreessen, a Silicon Valley venture capitalist who advises President Trump.

 

Experts point out that although DeepSeek's technology is less advanced than the AI ​​models developed by OpenAI or Alphabet (Google), it is still very advanced, considering that it was created using less advanced chips, fewer of them than American AI models, and some steps that Western researchers considered necessary to create AI models were omitted.

 

"DeepSeek declares that it has allocated approximately $6 million to train the new AI model, which took it several months, which is only a fraction of the amount and time that its US competitors had to spend to train the current AI models of major technology conglomerates. This information has caused some investors to question the validity of the previous, huge investment outlays on the development of AI technology, which at the same time and despite previous beliefs about the stagnation of Chinese competition in this area, was developed much cheaper and, it seems, similarly effectively," wrote XTB analysts.

 

Chamath Palihapitiya, a technology investor who used to sit on the board of Facebook, warned that the success of the DeepSeek model could undermine the sense of large investments in artificial intelligence made by American companies.

 

The success of DeepSeek, which has reached the level of the best AI models (from OpenAI or Google) with "insignificant" demand for graphics processors, shows that the restriction of chip exports planned by the US does not have to be a problem for Poland.

 

DeepSeek is being developed by Liang Wenfeng, a Chinese hedge fund tycoon known for his investments in artificial intelligence. On January 20, he met with Chinese President Xi Jinping and discussed how companies from the PRC can close the gap with the US in artificial intelligence research.

 

DeepSeek has already overtaken ChatGPT in terms of downloads in the Apple online store."

 

How is our personal and our company's data used to train ChatGPT? This can be avoided with the click of a button

 

"Artificial intelligence has become so naturally integrated into our everyday lives that many people, without even thinking, when a work or household issue arises, when they run out of ideas, open the chatbot ChatGPT, which is accessible to everyone, and its possibilities are practically limitless. But what do we know about the threats to the security of our confidential data? Experts say that there really are such.

Voldemaras Kadys, cybersecurity architect at NFQ Technologies, says that ChatGPT is most often associated with everyone as a tool for improving productivity.

“This artificial intelligence chatbot is used for personal and work purposes. It is important to note that in order to improve its quality and make it work even better, ChatGPT needs to be trained. We notice that newer and better versions of this tool are being released, which are trained on various data,” says V. Kadys.

According to him, first of all, these are synthetic data that are artificially generated, but efforts are made to make them look as real as possible, and then our personal data follows.

“Here is the bad news, because we enter everything there: personal problems, personal questions, as well as work problems, work questions. Some companies have ChatGPT business accounts, but most often everyone uses the free version, which is available to everyone,” says the interviewee.

According to him, the problem is that the privacy policy states that the data of those users who choose the free version that they enter can be used for training purposes.

“However, this can be solved, because the privacy policy also clearly states that if we do not want this, we can disable such an option, even in the free version.

All you need to do is select “do not train on my content” in your privacy settings,” the cybersecurity expert comments.

V. Kadys says that the situation is a little better for those with business accounts, because this button is automatically disabled there, but not everyone uses it.

“As a rule, this is not a cheap tool, because buying accounts for all employees so that everyone can freely use ChatGPT would be quite an expensive solution. So most people use the free version, and the conversation history can likely be used for artificial intelligence training purposes,” he warns.”

 

Kaip mūsų asmeniniai ir mūsų įmonių duomenys panaudojami „ChatGPT“ apmokyti? To išvengti galima vieno mygtuko paspaudimu

 

"Dirbtinis intelektas jau taip natūraliai įsiliejo į mūsų kasdienybę, kad daugelis net nesusimąstydami iškilus darbiniam ar buitiniam klausimui, pritrūkus idėjų, atsidaro visiems prieinamą pokalbių robotą „ChatGPT“, kurio galimybės – praktiškai beribės. Tačiau ką žinome apie ten slypinčias grėsmes mūsų konfidencialių duomenų saugumui? Ekspertai teigia, kad tokių tikrai yra.

„NFQ Technologies“ kibernetinio saugumo architektas Voldemaras Kadys teigia, kad „ChatGPT“ visiems dažniausiai asocijuojasi kaip įrankis produktyvumui pagerinti.

„Šis dirbtinio intelekto pokalbių robotas naudojamas asmeniniais ir darbiniais tikslais. Svarbu atkreipti dėmesį, kad norint, jog būtų pagerinta jo kokybė ir jis veiktų dar geriau, „ChatGPT“ reikia apmokyti. Pastebime, kad išleidžiamos vis naujesnės ir geresnės šio įrankio versijos, kurios yra apmokomos įvairiais duomenimis“, – sako V. Kadys.

Pasak jo, pirmiausia tai yra sintetiniai duomenys, kurie yra dirbtinai sugeneruoti, tačiau stengiamasi, kad kiek įmanoma atrodytų kaip tikri, o toliau seka mūsų asmeniniai duomenys.

„Čia ir yra blogoji žinia, nes ten mes suvedame viską: asmenines problemas, asmeninius klausimus, taip pat ir darbines problemas, darbinius klausimus. Kai kurios įmonės turi „ChatGPT“ verslo paskyras, tačiau dažniausiai visi naudojasi nemokama versija, kuri prieinama visiems“, – dėsto pašnekovas.

Anot jo, problema ta, kad privatumo politikoje parašyta, jog tų naudotojų, kurie renkasi nemokamą versiją, duomenys, kuriuos jie suveda, gali būti panaudoti apmokymo tikslais.

„Tačiau tai galima išspręsti, nes privatumo politikoje taip pat aiškiai parašyta, kad jeigu to nenorime, galime išjungti tokį pasirinkimą, net ir nemokamoje versijoje.

 

Tereikia privatumo nustatymuose pasirinkti „do not train on my content“ (liet. „nevykdyti mokymų pagal mano turinį“)“, – komentuoja kibernetinio saugumo ekspertas.

V. Kadys teigia, kad turintiems verslo paskyras situacija yra šiek tiek geresnė, nes ten šis mygtukas yra automatiškai išjungtas, tačiau ne visi tuo naudojasi.

„Kaip taisyklė, tai nėra pigus įrankis, nes nupirkti paskyras visiems darbuotojams, kad visi galėtų laisvai naudotis „ChatGPT“, būtų pakankamai brangus sprendimas. Taigi dauguma naudojasi nemokama versija, o pokalbio istorija, tikėtina, kad gali būti panaudota dirbtinio intelekto mokymo tikslais“, – perspėja jis.”