philipp flenker

The #llm sphere is moving so fast - a while back Claude was my favorite LLM, but these days I find that both Claude and Gemini have a strong tendency for sycophancy, overengineering and making mistakes.

Gemini is especially bad at this: Whichever problem I throw at it, it
a) congratulates me on having hit on a common issue
b) designs a solution that goes well beyond what I wanted and
c) stating that on top, "it's crucial to..." followed by something that is not true.

Gemini doesn't seem to believe that I am a capable person. When I asked it to show me how to implement a global handler for "Tab" in Obsidian, it showed me how to do something else, because surely I didn't mean to break things by globally handling "Tab" myself...?