You use Deepseek (not) like this

By Bert van Dijk Volunteer Apple.hcc.nl

The new free app Deepseek from China has become the most popular free iPhone and Android app in America in America. Through step-by-step reasoning, the app works much cheaper with only 10% of the usual computing power, because less power consumption and cheaper slower chips. As a result, the market value of AI chip manufacturer Nvidia fell on 1 day with no less than $ 589 billion! In this article you can read more about all the options and how you can easily use it for free. You can also read whether Deepseek is trained with knowledge from Chatgpt. Incidentally, it was only discovered that a vulnerability in Deepseek’s database led to exposure of sensitive information; Read more about this and also about the benefits and disadvantages of AI Local for better security.

You can do this with Deepseek
To answer questions, writing texts and programming code, Deepseek often performs better or comparable at much lower costs if the paid versions of Chatgpt, Claude and Gemini from Google. Within the iPhone app will probably also be a possibility to create images via Deepseek’s Janus model. The first results seem to be a little less regularly less than in Chatgpt and Grok within the free X app. With subjects that are sensitive in China, you sometimes get no answer or an answer that is strongly colored by how to think about it in China. Since January 20, the Deepseek-R1 language model has been running with 671 billion parameters. Six lighter open source versions are available via GitHub with 1.5 to 70 billion parameters. You can adjust these open source versions yourself and run locally with, for example, Ollama on a gaming PC or a recent Mac such as the Mac Mini with M4 chip that is very suitable with some extra internal memory.

(Text continues under the image)

This is how you use the Deepseek app for free
First of all an important tip and general advice with free AI tools: never enter personal or confidential data. Below is the step-by-step plan for using the free Deepseek app:

Step 1: Download Deepseek in the Apple App Store or via the Google Play Store for Android devices
Step 2: Do not use the app with your telephone number, but log in to iPhones and iPads via Apple with a hidden e-mail address. Due to the great interest in Deepseek in particular the web version Deepseek.com has been overloaded, but you may also be able to use it a little later via the Deepseek app.
Step 3: You can just ask questions in Dutch. Via the plus at the bottom you can add documents or a photo that you can ask a question about.
Step 4: With Deeply You can make Deepseek a little longer think for a better result. It is also useful that you can follow the thinking process.

Deepseek trained with knowledge from chatgpt
According to David Sacks, AI chef of the Trump government, Deepseek has trained on a large scale with knowledge from Chatgpt and are sufficient evidence for this. By Knowledge Distillation Can it learn a model through millions of questions from other models and thus suck up the knowledge and learn to reason.
Microsoft, which also cooperates with OpenAI, suspects that restrictions have been circumvented through several intermediate parties when using the OpenAI developer api. OpenAi will also investigate how Deepseek could be trained with the knowledge from their chatgpt models. A year ago, OpenAi indicated that it is impossible to train ‘Generative AI’ well without the use of intellectual property of third parties. Because they do not pay for that themselves, various content companies have brought lawsuits. So now it seems that one thief has now been robbed by another thief. Deepseek needs much less powerful chips, because the answer is made up of the answers of the most suitable experts (tired or mixture of experts) and MLA (Multi-Head Latent ation), which requires much less memory. In the last V3 version, Load Balancing and an automated smart training in 2788 hours for less than 6 million dollars were made many times more efficient. You will find more extensive information about these efficiency improvements in this Deepseek FAQ.
Because with certain questions about the ‘Tankman’ on Tiananmen square, Sensitive Information Sleeps through, that is a clear indication that the answers from Deepseek are not yet entirely from a dataset that is fully constructed in China. Apart from this, it is in any case an open source model that performs exceptionally well. It is also favorable that many companies can also use unlimited commercial commercial in-house due to the MIT Open Source license. For example, they do not have to be afraid that their data will end up at one of the American Big Tech companies.