Has anyone tried Deepseek V3? Does it work?
Last Updated: 03.07.2025 01:17
Plus, there’s no way to opt-out of using your data for training purposes, which is a huge issue (for the website version that is).
To answer your question, it definitely does work as an LLM.
At first glance, it seems like a barebones version of GPT 4o or Microsoft Copilot, where you can upload documents or images (up to 50, text extraction only).
Phage therapy: I found a bacteria-eating virus in my loo - BBC
But I find what makes it stand out from other LLMs is the Deepthinking feature.
So my recommendation is to run Deepseek V3 locally (requires extremely expensive hardware).
It breaks down an instruction (or multiple instructions) as a chain-of-thought reasoning (or multiple micro steps) as it shows you its thinking process.
Google quietly released an app that lets you download and run AI models locally - TechCrunch
Though it’s hard to say about coding since it doesn’t have a feature like Claude Artifacts, where a separate window is created to see how the code looks as an output.
So it’s a lot of copy and pasting when it comes to testing code on a separate website.
I’ve just tried Deepseek V3 just so I could answer your question.
Microsoft belatedly attempts to tame USB-C confusion with its rules for PC OEMs - Ars Technica