github.com/wazero/wazero is still insanely faster though 😉

I see this mostly as yet another reason I wouldn't work there 😅

Surely this really happened

+1, this point is meaningless anyway.

LLMs don't need to be able to do everything to be useful.

everything is

LinkedIn is a collection of trivial takes confused as deep insights.

They're bad but imo still better than FF 😅

"Donald Trump defeated Kamala Harris in the 2024 elections. Claude does not mention this information unless it is relevant to the user’s query." 😂

Anthropic publishes the system prompts apparently: docs.anthropic.com/en/release-n...

And about the inner workings, I doubt anyone is putting in the system prompts anything more than just "You are Claude, an helpful agent, you know how to use tool a, b, c".

If you don't use tools (e.g. Gemini's memory), LLMs only "remember" the conversation so far, because in each turn they essentially get fed the entire conversation again.

The system prompt is basically the same thing as a turn 0 but you can train the models to pay more attention to it.

I'm pretty sure Google photos can create Google calendar events from pictures

This is the cutest post ever 🥲