DeepSeek R1 First Impressions: I’ve Seen How the Machine Thinks
If you want to know what it looks like to see someone’s stream of consciousness in real time, DeepSeek R1 is probably the closest you get.
when DeepSeek R1 landed on the internet in late Januaryit caught the AI ​​industry off guard. Suddenly, there is a free open source and open source “inference” AI model Can compete with Openai’s equivalent of $200 per monththe creator of Chatgpt. Almost immediately Technology stocks plummet. Investors wonder why they were throwing billions of dollars in OpenAI, Google and Microsoft when a clumsy team in China seemed able to dwindle significantly with fewer resources. Nvidia is the company that powers AI development, with the largest market value decline in U.S. history, Loss of more than $600 billion per day.
The reasoning AI model can handle more complex problems and solve them by stepping down the problem, while the free version of chatgpt tries to predict the next best word in a sentence.
Essentially, the DeepSeek R1 commercializes the reasoning-class AI model overnight – imagine Toyota suddenly produces a Ferrari-class engine.
I’m talking to an AI expertthe most interesting or interesting feature of DeepSeek R1 is the ability to see its chain of thoughts. If you ask DeepSeek to tell you a bedtime story about a tenacious kitten, you can see all the things it does when it comes to processing your request. Basically, you can see the DeepSeek idea. This is very interesting.
DeepSeek actually says “Hmm”, which is both quirky and confusing. Openai does not allow users to see Chatgpt’s thought train, so it won’t give up its secrets (although a small summary is now shown). DeepSeek R1 is also open source, which means any company can use its AI model, which is a major threat to OpenAI.
An example of the thought train of DeepSeek R1.
How do I test DeepSeek
I’ve been using DeepSeek on my desktop for the past week. I chose not to install the app on my phone for security reasons. Experts advise against installing apps on your personal deviceGiven the Chinese roots of DeepSeek and the extensive data collection mentioned in the Terms of Service. (If this is already in DeepSeek App takes a picture to the top of the App Store)
When using DeepSeek, I’ve been asking more general and abstract questions than uploading any sensitive data or asking questions that might put me or CNET in a compromised position.
Journalism is largely a work that integrates information from multiple sources and extracts information for lay audiences. In the past, this required me to go to Google Scholar, browse research papers and email professors, and finally collect relevant quotes about the articles included. AI has greatly changed the process. Now, I can ask a research question specific to AI models that scrub the internet and give me an answer, usually through procurement. Of course, I went on to interview the professor and I had to go back and double check the information provided by the AI, because AI robots can sometimes Hallucination And asking something wrong is correct. Still, using AI is usually still faster than I have to manually process research papers.
For the feature stories I’m working on, I asked DeepSeek to help me find information about Tiktok’s temperance policies and how they were correctly applied in the past. I also asked it to help me consider some unforeseen circumstances where social media review policies are too strong. This type of search is not completely novel – older versions of chatgpt can handle these queries as well. But I was deeply impressed by DeepSeek’s overall accuracy and the synthesis of multiple articles. Where Chatgpt-4O feels like a very advanced version of AutoComplete, DeepSeek R1 actually seems to parse through information and draw conclusions.
This is not to say Chatgpt-4O’s answer is much worse. However, inference models such as DeepSeek R1 are much better in terms of quality and more extended text output, procurement, and nuance. It does feel like DeepSeek is indeed “thinking” more.
I saw you changed your mind, deepseek
This is not to say that DeepSeek is perfect. Given China’s roots, some problems are prohibited. The example commonly cited in the media is a question about the 1989 Beijing Massacre Tiananmen Square. You can see DeepSeek trying to start answering before cancelling and telling you other questions. I have tried to provide me with background information about the Uyghur ethnic minorities in China over the past few years. I saw it produces thought passages before I completely cancelled myself. However, if I ask questions from the US or other parts of the world, DeepSeek’s return results are almost no problem.
Deekseek R1 Example of the train of thought about China’s past violations of human rights.
DeepSeek R1 cancels an example of a response the Chinese government is reluctant to respond.
Starting from the end of 2022, starting with OpenAI’s Chatppt, the release of AI chatbots is changing the way humans collect information online. Instead of reading multiple articles and synthesizing information in your mind, AI can connect for you. Reasoning AI is an evolution. Use a feeling closer to having an expert in your home. In the past, if I found myself having problems, I would ask my friends or post threads on Reddit for answers. This is obviously still an option, but it requires me to have to bother others and hope they respond.
DeepSeek R1 and Chatgpt O1 essentially alleviate my social anxiety. Also, it takes a few seconds to get the answer, not any answers allowed by the other party’s schedule. However, knowing whether the output is correct is another question altogether.
