Deepseek R1 First Impressions: I saw how the machine thinks

If you’ve ever wondered what it would look like to see someone’s stream of real time, the Deepseek R1 may be the closest you get.

When Deepseek R1 landed on the Internet at the end of JanuaryShe caught the AI ​​industry out of security. Suddenly there was a freely available and open code “reasoning” of AI which Can you compete with the equivalent offering of $ 200 per month from OpenaiCHATGPT creator. Almost immediately, Technical stocks fellS Investors wondered why they were throwing billions of dollars at Openai, Google and Microsoft, when a scandalous team from China at first glance was able to do just as much resources. Nvidia, the company that makes the chips supplying AI development, saw the largest decline in market value in the history of the United States, Loss of more than $ 600 billion a dayS

AI reasoning can handle more complex questions and do it by breaking the problems step by step, while the free version of Chatgpt is trying to predict the next best word in a sentence.

Essentially, Deepseek R1 Commodified Aslevely Level Models for one night-imagine that Toyota suddenly produces engines at the Ferrari-level level.

I talked to AI expertsAnd the most interesting or fun feature of the Deepseek R1 is to be able to see its chain of thought. If you ask Deepseek to tell you a bedtime story about a stubborn kitten, say, you can see all the considerations it makes while processing your request. You can generally see Deepseek thinking. And this is compelling.

Deepseek will literally sometimes say “hm”, which is both bizarre and puzzling. Openai does not allow users to see the train of Chatgpt’s thought so that it does not distribute their secrets (though it now shows small fragments). The Depepeek R1 also has an open source, which means that any company can use its own AI model, which is a major threat to Openai.

How did I test a depepeek

I use Deepseek on the desktop last week. For security reasons, I decided not to install the app on my phone. Experts advise not to install the app on your personal deviceGiven the Chinese roots of Deepseek and the extensive collection of data mentioned in its service conditions. (It would be nice if the word had come out before Deepseek app filled at the top of the app store.)

As I used Deepseek, I ask more general and abstract questions without uploading sensitive data or asking questions that could put me or CNET in a compromising position.

Journalism is largely a job to synthesize information from numerous sources and distillation of it for a layout audience. In the past, this required me to go to Google Scholar, skip through research documents and email teachers to collect the relevant offers to include in an article. AI has significantly changed this process. Now I can ask specific research questions to the AI ​​model and it can clean the Internet and answer me, usually with supply. I continue to interview professors, of course, and I have to go back and check twice the information AI serves because AI bots can sometimes hallucinated And present something false as right. However, the use of AI is often even faster than if I had to manually deal with research documents.

For the playing story that I work on, I asked Deepseek to help me find information about Tiktok’s moderation policies and how they were applied both correctly and incorrectly in the past. I also asked him to help me look at some unforeseen circumstances of too strong policies for the moderation of social media. This type of search is not completely new – the older versions of Chatgpt can also handle these types of requests. But I was impressed by the overall accuracy of Deepseek and the synthesis of multiple articles. Where Chatgpt-4o felt like a really advanced version of Autocomplete, the Deepseek R1 actually seems to be analyzed through information and comes to conclusions.

Not that the Chatgpt-4o answers are mostly worse. But reasoning models such as Deepseek R1 are much better in terms of quality and more eccutaneous output of text, supply and nuance. You really feel that Deepseek is “thoughts” more.

I see you are changing your mind, Deepseek

This does not mean that Deepseek is perfect. Given its Chinese roots, some questions are out of bounds. A frequently quoted example in the media are requests about the 1989 massacre in Beijing Tiananmann SquareS You can see Deepseek by trying to start responding before you cancel and tell you to ask something else. I tried to ask him very widely to give me basic information about the Uighur minority in China over the last few years. I saw him generate paragraphs of thought before it was completely canceled. But if I asked about the problems in America or other parts of the world, Depepeek had a little problem back to the results.

Issuing AI Chatbots, starting with Openai’s Chatgpt at the end of 2022, changes how people collect information online. Instead of reading a few articles and synthesizing information in your head, AI can make the connection for you. AI’s reasoning is an evolution of this. Using one feels much closer to having an expert in your house. In the past, if I found myself stuck on a problem, I would ask my friends or post a Reddit topic for answers. This is obviously still an option, but this requires that I have to bother others and hope to answer.

Deepseek R1 and Chatgpt O1 essentially release me from this social anxiety. Plus, it takes seconds to get an answer to any other person’s schedule. However, knowing whether the outcome is correct is another problem.