Jinwoo Je

LLMs (like Claude) are surprisingly good at giving movie/music recommendations. Provide a list of your likes and ask for suggestions. https://claude.ai/share/ee321e18-ca97-4aa4-93e9-6b1c7c74aa03
Decided to set up a budget specifically for testing, interacting with, and learning more about LLMs, given the exponential productivity gains and flattened learning curve.
Pro subscriptions for chat interfaces (Claude/Gemini/ChatGPT) remain cost-effective compared to API credits, especially considering beneficial features like chat history, artifacts, canvas, and voice interaction.
Tested the newly released Sonnet 3.7 model. Wasn't significantly more impressed given the high bar of 3.5 Sonnet. Rate limits and slow token generation are bottlenecks compared to faster models. Reallocated the subscription cost ($20/month) towards API credits instead.
Switched to testing Gemini Pro. The latest Flash model is impressively fast, and the multi-modal capabilities (especially voice input) are growing on me.
Exploring MCP tools to improve workflows. High potential, but haven't yet reliably integrated them. Integration with note-taking apps (like UpNote, which currently lacks MCP support) would be valuable for LLM I/O. Tested Todoist integrations, but found execution unreliable. Continuing to monitor and test.
Expect LLMs to cause an exponential increase in the volume of digital data.
Aligned with using APIs directly, I'm testing CLI-based AI tools like aider and Claude Code.
Prioritizing Markdown for all writing. This aims to future-proof content, avoid app lock-in, and improve LLM interaction.