Mastodon offers a feature to set up filters for excluding posts containing specific keywords. This week, I decided to use it on my Fosstodon.org account to filter out posts related to U.S. politics, Elon Musk, and LGBTQ topics. It's not that I'm uninterested or unconcerned about these issues, but I joined Fosstodon for its focus on tech and nerdy hobbies. I follow people who primarily post about these topics.

However, as my follow list grows and the instance expands, I’ve noticed an increasing number of posts on these topics. Unfortunately, the tone of such posts is often angry, condescending, or outright abusive—something I’d rather not mix with my hobbies and interests. For me, Twitter, Bluesky, and news channels are already sufficient sources for staying informed about these issues.

Since setting up the filters a few days ago, I’ve felt much less triggered and more at ease. The takeaway? Keep your hobbies separate from politics and other social issues unless discussing them is your hobby. Mixing the two can drain the joy from what should be a pleasurable escape.

For the past two Sundays, I have been recording the church sermons with the Otter.ai app, hoping to archive the audio and the transcribed text for future use. However, Otter only allows 30 minutes of transcription in its free plan, which is usually 15 minutes short on a typical Sunday. So, today, I asked Fediverse if there is a free or low-budget solution to my problem, and the suggestions were to use a locally installed Whisper model from OpenAI and use it to transcribe the audio recording either via script or through some off-the-shelf GUI. I tried both and settled with a cross-platform app called Vibe that uses the Whisper model to transcribe audio files locally and also has features for summary and context generation through APIs to other popular models. I used it on two recordings and am pleased with the results. It took some time to transcribe a 30 to 45-minute recording, mainly because I have a very old, slow computer without GPU support. However, Vibe provided the transcribed text in several formats, including .srt, .txt, .html, and .pdf. I check if the .srt and audio files are in sync using the VLC and Celluloid media players on Linux Mint. I used the transcribed text to generate a summary using ChatGPT. For now, this workflow serves the purpose.

Today, I read this beautifully written article, "Planning for an Uncertain Future" (link below), by Nicholas Johnson, on his Gemini website. I may not agree with all the details that he mentioned, which sounded very confident and methodological, but I do agree with the overall message that too much future planning at the expense of current joy and life fulfillment is futile. So, do today what you can do today to live a better life than worrying or overplanning for your future.

Note: you will need a Gemini browser to open the link: gemini://nicholasjohnson.ch/2024/12/09/planning-for-an-uncertain-future/

I am working with a data frame with 53 million rows and 28 columns, which I stored as a .feather binary file. On the disk, it's around 5 GB. Fortunately, I am working on an HPC and have ample RAM to load the entire data frame on the memory. However, working with such a huge file is slow and cumbersome.

Happy Thanksgiving, USA, and every other country that celebrates it. We had an Indian-American mixed family lunch today, and we enjoyed the variety of food and spending time with the family after a long time. One of my colleagues coined a new term for our kind of lunch today: "Curkey." That's a mix of curry and turkey. We all loved it.