Back
Research
Load previous page…
🤖 Updated some scripts to use Ollama's latest structured output with Llama 3.3 (latest) and fell back to Llama 3.2. I drop from >1 minute with 3.3 down to 2 to 7 seconds per request with 3.2. I can't see a difference in the results. For small projects 3.2 is the better path. 
#research
🤖 Tried out Llama 3.3 and the latest Ollama client for what feels like flawless local tool calling. #research
🤖 I realized I didn't need to chain multiple tool calls together for my Calendar Agent to work. I can have my create check and then stop overthinking it. #research
🤖 I hit a wall where my Calendar Agent only tries to create new events on March 28th, 2024 #research
🤖 my CalendarAgent can now read and write to my calendar. I haven't quite figured out how to make it smart enough to know when to look for what's already on my calendar. i was hoping tool calling would work a little better. #research
🤖 🗓 Wrote some code to connect to my Calendars and now I'm teaching an LLM to read and write to it #research
worked up a slightly different engine for my files-to-claud-xml project to make the internal data structure easier to build other tools with. #research
Used Claude to write a Django management command to flatten my templates and pull out third party templates that might be used #research
🚜 Finally got around to switching a few APIs over to use claude-3-5-haiku-20241022 and I cut my bill down by 80 to 90%. Not bad and I'm not seeing a noticeable difference in quality. #research
🎙 met with a former ED for an open source project and had a wonderful conversation about governance and all-the-things. It was quite lovely. #research
🤖 Finally remembered the jina.ai address (been looking for weeks) and made a quick Agent to pull in remote pages. #research