Artificial intelligence-powered chatbot / large language model is a new child on the block, and it is threatening to blow the rest out of the water. Meet Dipsec, with a part of the budget developed by a Hangzo-based research lab (if you believe rumors) are used to make chatgates, Gemini, Cloud AI and others made by the United States-based computer lab in the United States. .
And the latest Prasad – Deepsek V3, a 671 billion parameters, ‘Mixing of experts’ model; And Deepsek R1, an advanced logic model that uses AI, is possibly better than the 01 of OpenaiI – has underlined its position as a possible heavyweight financial and technical disruptive in the region.
How disruptive is this?
On Monday, this app, Dipsek V3 model, is now the top downloaded app on the Apple Store in the US; That sink in … a Chinese-developed chatbot is now the most downloaded app in America.
And that disintegration, even if seen as a ‘potential’ at this time, has raised doubts about how well some American tech companies have invested the Arabs pledged towards AI development.
Either way, the quality and cost efficiency of the models of Deepsac has flipped this story; Even if, in the long run, it is a special Chinese model flop, that it was developed with a fraction of financial and technical resources available to firms in the west, an eye -opening.
Then, how disruptive is it?
Well, the creators of Deepsek said last month The V3 model requires less than $ 6 million (Although critics say that connecting costs to the stages of earlier growth can lead to the final costs in computing power from NVidia’s H800 chips). Bernstein analyst Stacey Rasgan told Reuters, “Deepsek actually built Openai for $ 5 million?
But break the available financials and it becomes quite remarkable.
Openai’s 01 fee $ 15 per million input tokens.
Deepsek’s R1 $ .55 per million input charge tokens.
Pricing, therefore, blows up the competition completely.
And, based on cases of end-use, It is believed that the deepsake is 20 to 50 times cheaper, and efficientCompared to 01 models of Openai. In fact, logical logical test score results are staggering; Deepsek chats and clouds make AI better than seven to 14 percent.
A popular online community for software developers dev.to said that it scored 92 percent in complex, problem-solution tasks compared to 78 percent by GPT-4.
Input tokens, by the way, see information units as a quick or part of the question. These are basically that the model needs to analyze or understand the context of query or instructions.
For reference, openiI is considered to spend $ 5 billion every year to develop its model.
Therefore, even though the critics (see above) of Deepsek are correct, it is still a fraction of the costs of Openai.
This translates, as the company’s boss Sam Altman said, in a significant computing capabilities, but at least for the Deepsek model, there is a eye-razor to provide a lot of processing power on its relatively shastring budget.
And Mr. Altman admitted that, calling the R1 model “very impressive”.
Google Boss Beautiful Pichai moved one step forward, telling CNBC in Davos, “I think we should take development from China very seriously.” And US President Donald Trump has called a “wake-up” for American tech and computing giants.
And there are hundreds of billions of dollars that American companies have lost in technical shares between a route this week; For example, chip-producer Nvidia, more than $ 600 billion and the Tech-Rich NasDaq index ended over three percent on Monday, with another decline likely to decline based on the expected income reports of AI veteran Meta and Microsoft.
For reference, both meta and Microsoft have their AI models, at the forefront of which are Lama and Copilot; The former is an LLM that was first released in February 2023 and later now has an integrated feature in various Microsoft 365 applications, such as MS Word and Excel.
While neither, of course, at the same technical levels of Openai or Chatgpt, Meta and MS have invested billions in AI and LLM projects, which are both in the US and abroad. For example, some analysts believe that Big US Cloud companies will spend $ 250 billion on AI Infrastructure alone this year.
But the one that actually makes Dipsek special is higher than the cost and technology.
It is, unlike its rivals, it is really open source.
R1 code is completely open to the public under MIT licenseWhich is an permissible software license that allows users to use, modify and distribute software with certain restrictions.
This means that you can download it, use it commercially without fees, change its architecture, and integrate it in any of your existing systems.
Deepsek is also faster than GPT 4, more practical and, according to many experts, even understands regional idioms and cultural references than its western counterparts.
There is a lot of thought.
How, for example, Deepsek affects diplomatic and military relations between China and America (and India, really), and what are the moral problems with the open-source AI model?
But what is undisputed is that the lamp of China is a disruptive, financially and technical. And experts believe that China has now jumped-18 to six months behind the state-of-the-art AI model developed in America.
With input from agencies
NDTV is now available on WhatsApp channels. Click on the link to get all the latest updates from NDTV on your chat.