Wait, You Can Do That? How (and Why) to Install a Neural Network on Your Computer

Most neural networks currently work online, but some models can be downloaded and run locally on your own computer, even without internet access. This is called local AI deployment. It is legal: developers publish model weights for download, though each model has its own license. Running locally means your data stays on your machine, you can work offline, and you avoid API fees. However, local models are generally less powerful than cloud ones and may require significant RAM or a good GPU. Popular models for local use include DeepSeek, Llama from Meta, Qwen from Alibaba, Gemma from Google, Mistral, and Microsoft's Phi. For installation, programs like LM Studio simplify the process. As an example, to run DeepSeek with LM Studio, you download the software, search for DeepSeek, choose a quantized version (e.g., Q4_K_M), and download it. System requirements vary: DeepSeek 8B needs at least 8 GB RAM and 5-6 GB disk, 14B wants 16 GB RAM, and 32B requires 32 GB RAM. ChatGPT cannot be downloaded; OpenAI does not release weights. After installation, internet is only needed for updates or new models. A powerful GPU isn't mandatory; CPU works but is slower. Quantization reduces model size and requirements at some quality cost.