The Nerd Show | News Videos and Reviews

Bash Functions. It's a crime not to use them.

Fixing annoyances. Bash functions give us a way to dramatically improve the Linux terminal experience. Unfortunately, only a select few take advantage of them. Functions, as the name suggests, provide a little more functionality than the aliases we blogged about earlier. We can use them to simplify our workflow by combining powerful operations into a single command.

Functions disappear. The problem is, functions are no longer available after a reboot. In this post, we will

Happy New Year! It's time to alias your aliases.

The title is not quite correct. What we are going to do is create a function to improve Linux’s alias command. After that, we will use it to upgrade our desktop with a new search hotkey.

Testing gpt-oss-20B off-grid AI

The test. We tested a popular Open-source AI model on our laptop, an older Maingear Vector Pro with 32GB DDR4 and Nvidia RTX 3070 with 8GB VRAM and an AMD Ryzen 9 5900HX.

The results were pretty impressive for a model that's a little bigger than fits in VRAM comfortably. It took some trial and error to achieve the best generation time of 33 tokens per second. Which is a much better than the 7 tokens/sec. obtained with default options, and a lot faster than normal people can read.

To use these larger models we had to compile llama.cpp with CUDA enabled. And tell CUDA to use unified memory so that if the GPU becomes full it will spill over into RAM instead of crashing:

export GGML_CUDA_ENABLE_UNIFIED_MEMORY=1

Command. Here is the command we used that achieved the best results. Changing other parameters like --n-cpu-moe, -ub, or -c didn't seem to make any difference, so we went with the max context size (-c 0). This enables it to generate reams of text with fairly good speed on our portable AI box. The optimal number of graphic layers (-ngl) turned out to be around 14-16 40 8GB VRAM.

llama-server -c 0 -ngl 16 --jinja -hf DavidAU/Openai_gpt-oss-20b-NEO-GGUF:MXFP4_MOE4

Prompt. We asked the model to, "Give an example of a little-known task you can do that is very useful."

Decentralized, (& Simple?) Alternatives to YouTube

Setting up a curated torrent hub in 2025 is often achieved by combining a private indexer with a modern, self-hosted frontend. This allows you to hand-pick content while maintaining the speed and decentralization of the BitTorrent protocol. [This might not be the best way, but it is decentralized and simple]

1. Modern Self-Hosted Torrent "Hubs"

AI wonder. How does llama.cpp compress LLMs?

An alias makes everything easier.

No we're not talking about using a fake name.

Alias is a tool for the Linux terminal that allows users to combine several oft-used commands into one command that's easy to remember, and even easier to type. They are typically one liners. Any line of commands that is too short to make into a script or program in its own right can be an alias. Here are a few recent examples.

Limit Firefox's GPU consumption.

Five quick steps to boost StackOverflow reputation.

To increase your reputation on StackOverflow, follow these five quick steps: Yep, there's nothing new here. This rapid-transit route to StackOverflow cred concentrates on the basics.

Step 1: Participate in the Community

Contribute to the community by answering questions and engaging with others. This will help you build relationships and establish yourself as a knowledgeable expert in your field.

Best Free Open Source Tax Software for 2025

As tax season approaches, finding the right software to help you navigate your tax preparation can be a daunting task. In 2025, several free open source tax software options have emerged or improved, providing robust solutions for individuals and businesses alike. Here’s a look at some of the best available options.

1. TurboTax Open Source Edition

TurboTax has introduced an open source edition of their popular software, allowing users to download, modify, and redistribute the program for free. This version retains the user-friendly interface of the commercial version while offering the flexibility of open source.

With AI on every corner making wild claims about their performance, it can be difficult to cut through the hype. Here are some steps that you can take to familiarize yourself with the various commercial and open source models, their performance, resource requirements, tuning options, availability, use cases, and limitations.

Subscribe to this feed. sub