DeepSeek shocks world with latest upgrade! Valuable lessons emerge!

DeepSeek R1 model has just completed a minor version trial upgrade, that is, DeepSeek R1-V2! You can go to the official website, app, and mini program, for testing. The API and usage remains unchanged. 

DeepSeek just dropped a surprise R1 update that's sending shockwaves through the AI community - this "minor trial upgrade" might be the game-changer we've been waiting for! DeepSeek completed what it described as a "minor trial upgrade", and told users they can start testing it. What's more: DeepSeek beats Claude 4 and Gemini 2.5!

A voice dictation app allows you to be yourself. The chain-of-thought behavior appears to have changed significantly. It also appears to be quite quick.

Bijan Bowen, technologist, said DeepSeek has released a new version DeepSeek R1-0528, which is a very good model update. There is a browser OS emulator, as shown below.

You can double click on desktop icons to open applications. Use the taskbar to manage open windows. To use the OS emulator, you can double click any desktop icon, or use the Start menu to launch applications. You can move windows around, minimize them to taskbar, and switch between applications.

Do take note that this Chinese AI startup that already triggered a $1-trillion market sell-off with their original R1 model, is clearly not slowing down. This could be the catalyst that reshapes the entire AI landscape once again!

An example shown was making a Python game that allows a user to crash test a car with some soft body physics. There are instructions regarding playing the game. There is car crash simulation as well. Car deforms are based on simulation speed.

DeepSeek's infrastructure-level capabilities continue to be among the most under estimated. DeepSeek had started with fewer GPUs, and even had to use downgraded hardware like A800 and H20. Despite that, they pushed limits of the existing GPUs to an extreme, with optimization results that are more efficient than even Nvidia's official methods.

It also implies that should the broader industry adopt DeepSeek's optimization strategy, the existing GPUs could still yield significant performance gains.