0600-1000:
- Woke up to the news of a small mice on our kitchen. We were suspecting something was roaming our house when we found some traces of small droppings on the kitchen. Today the source has been identified. The problem is how did it end up on our kitchen and where did it come from.
- Made two eggs and avocado after waking up at around 0800 as my wife has gone for her work. Ate the breakfast with milk
- Today I have plans for learning about skills to be used with LLMs, and one fascinating use case of AI is the Google Gemma 4 2B params one, which works flawlessly on a iPhone, and is pretty powerful, with the examples showcasing its ability to call a javascript function. That could be very useful in the future to host my own local llm and work on things. Also, I am trying to learn skills that help me with this AI boom, even though I might not be able to land a job with them, but I am hoping to extract useful information from the existing AIs. Also discussed that on my recent thought 0002 AI and workforce.
1000-1400:
- Looked up the Gemma skills, and Claude skills as well. The code itself looks easy.
- On further dive, the skills have a specification guidelines, which determines how the skills should be organised. AgentSkills is where I should first visit if I were to create any skill, that is cross compatible with all AI models.
1400-1800:
- I had no reason to pursue the research, as the potential of the Google AI edge is high, I can use the AI model myself without token usage issue and can create cool stuff, but I dont have any idea what to do.
- Few ideas that dabbled on my head. One was to use the TFL api to get to where I want, but I feel like it might not be as useful as to use a third party app like Citymapper, which is convenient as well as efficient. Another was for myself to gather latest news about stocks I follow and summarize them, also possible with scraping google news and yahoo finance, but I didnt want to work on it today.
- Did some chores, and sat down for contenplation 🌼.
1800-2200
- Made dinner. Roasted chicken with veggies all roasted on air fryer tasted pretty good and was a healthy option than most.
- Lurked around on the internet to find tools like Defuddle, which apparantly also has an api call option, it directly converts webpages into markdown files compatible with Obsidian as well. Apparently, it is the tool used by Obsidian Web Clipper, an official browser extension from Obsidian itself.
2200-2400:
- I asked notion AI to write a story about Genie and wishes, based on my note 0012 What would be the wish about same topic, it cooked hard. I’m not sure which model it used, but it generated an absolute unit of a short story: 0002 Genie on a bottle.