Join daily and weekly newsletters to obtain the latest updates and exclusive content to cover the leading artificial intelligence in the industry. Learn more
Just one week ago-on January 20, 2025-Chinese Deepseek AI launched a new start Artificial Intelligence Model is called R1 This may be initially mistaken in one of the constantly increasing fans of almost switched competitors that appeared since Openai for the first time Chatgpt (supported by its GPT-3.5, at first) more than two years ago.
But this soon proved that it is unfounded, and the Deepseek application for the mobile phone at this short time The Apple App Store Plans in the United States were spent to evacuate Chatgpt for first place And cause Correcting the huge market with stock investors ’dumping In previously hot computer chips like NVIDIA, which GPS processing units (GPUS) were in a large demand for use in a huge superior infection to train new artificial intelligence models and serve them on customers continuously (a method known as “inference”.
Mark Anderson, capitalist capitalism, echoing the feelings of other technology workers, Books on the social network x last night“Deepseek R1 is a moment of Sputnik of artificial intelligence,” compared to the axis October 1957 The first artificial satellite launch in history, Sputnik 1, Through the Soviet Union, which raised the “space race” between that country and the United States to control space travel.
Sputnik launched the United States to invest heavily in searching and developing spacecraft and missiles. Although it is not an ideal likeness-there was no heavy investment to create Deepseek-R1, quite the opposite (more in this below)-it seems that it indicates a major turning point in the global artificial intelligence market, as is the case for the first time that has become The artificial intelligence product of China is the most popular in the world.
But before we jump on the noise train, let’s down and study reality. As a person who used Chatgpt from Openai on a large scale-on both web and mobile platforms-and follow the progress of artificial intelligence closely, I think that although Deepseek-R1 achievements are worth noting, it is not time to refuse Chatgpt or US AI yet. Please note, their salaries are not paid by Openai to say this – I haven’t got money from the company or planned.
What Deepseek-R1 do well
Deepsek-R1 is part of a new generation of large “thinking” models that make more user information: they are reflected in their analysis during a response production, in an attempt to capture errors before serving them to the user.
Deepseek-R1 or exceeds Openai’s thinking model, O1, which was released in September 2024 at first only for Chatgpt Plus and PRO users, in several areas.
For example, on the Math-500 standard, which evaluates the solution of mathematical problems at the school level, Deepseek-R1 has achieved a 97.3 % resolution, slightly outperformed Openai O1’s 96.4 %. In terms of coding capabilities, DeepSeek-R1 recorded 49.2 % on the Swe-Bused standard, hiding from Openai O1’s 48.9 %.
Moreover, financially, Deepseek-R1 provides large cost savings. The model has been developed with an investment of less than $ 6 million, which is a small part of the expenses – Introducing multiple billions – It is associated with reporting with training models such as Openai’s O1.
Deepseek was mainly forced to become more efficient with rare and the elderly Graphics processing units thanks to the US export restrictions on technology sales to China. In addition, Deepseek provides API’s arrival at $ 0.14 per million icons, which greatly reduces the Openai rate of $ 7.50 per million icons.
DEPSEK-R1, costs and equivalent performance of the highest AI model in the United States, Silicon Valley and the broader business community about what appears to be a integration of the artificial intelligence market, geological booked, and the economy known in artificial intelligence model training.
While Deepseek’s gains are revolutionary, the pendulum swings very far towards it now
No one denying that the effectiveness of Deepseek-R1 is a great achievement. But let’s not forget that Deepseek itself owes a lot of its success for us, Amnesty International innovations, and to return to the structure of the 2017 initial transformers developed by Google Ai (which began the entire LLM madness).
Deepseek-R1 has been trained in the questions and answers of the artificial data data, and according to the paper released by its researchers, the “Deepseek-V3 Data Data Group”, which was supervised, which was found, which was found to be Many indicators created using the GPT-4O model from Openai himself!
It seems very clear to say that without GPT-4O to provide these data, and without the Openai version of the first O1 O1 commercial thinking in September 2024, which created the category, Deepseek-R1 will not be almost almost present.
Moreover, the success of Openai requires huge amounts of graphics processing unit resources, which paves the way for subscriptions that Deepseek has benefited undoubtedly. The current investor feels panic around Chip and AI by artificial intelligence companies.
ChatgPT capabilities and image generation are still very important and valuable in the workplace and Deepsek-R1 have no any yet
Although Deepseek-R1 admired her thinking “a series of” a series of thought “-a kind of awareness of awareness where the model displays the text during his analysis of the user and seeks to answer it-and efficiency in the progress of text-based action and mathematics, it lacks many features that Chatgpt makes a more powerful and diverse tool today.
There are no possibilities for generating images or vision
The official Deepseek-R1 and the mobile application allows users to download photos and file attachments. However, they can only extract the text from them using visual letters (OCR), one of the oldest computing technologies (It dates back to 1959).
This diminishes compared to the vision capabilities in ChatGPT. The user can download images without any text at all and to have a picture cargo analysis, or describe it, or provide more information based on what he sees and the user text claims.
ChatGPT allows users to download images and can analyze visual materials and provide detailed visions or implementable advice. For example, when you need guidelines about repairing my bike or maintaining my air conditioning unit, ChatgPT has proven to process images that they are invaluable. Deepsek-R1 simply cannot be done yet. See below for a visual comparison:
There is no images generation
The absence of obstetric images is other major restrictions. As a person who generates AI’s images frequently using ChatGPT (such as the head of this article) supported by DALL · E3 from Openai, the ability to create detailed and stylistic images with Chatgpt is the game change.
This feature is necessary for many creative and vocational workflows, and Deepseek has not been proven after similar jobs, although the company has today released an open source vision form, Yanos ProWhich it says is outperforming Dall · E3, stable spread 3 and other models of generation in industry on third -party standards.
No sound putting
Deepseek-R1 also lacks the audio reaction mode, a feature that has become of increasing importance to access and comfort. Chatgpt audio mode allows natural interactions and conversation, making it an interesting option to use without using hands or for users with different needs to reach.
Be enthusiastic about the future Deepseek capabilities – but also be careful of its challenges
Yes, DeepSeek-R1 can add-the potential of sound and vision in the future. But doing so is not easy.
Integration of images, vision analysis, and vocal capabilities requires significant development resources, and it is irony that many high -performance graphics processing units are the same that investors now say. Spreading these features effectively and in an easy -to -use manner is another challenge.
Deepsek-R1 achievements are impressive and indicate a promising shift in the global scene of Amnesty International. However, it is important to keep excitement in choice. Currently, Chatgpt is still the best and most capable product, and offers a range of features that cannot match Deepseek. Let’s appreciate developments with the constant restrictions and importance of innovation and investment from artificial intelligence.
