Please fix the models abilities to interact and even read the notebooks, it’s been way too fucking long, even Claude is so buggy @cursor_ai.

Why does Netflix not have an ai assistant for finding things to watch? @netflix.

Why does maven on audible not have access to my reading history? @audible_com.

"To form a clear idea of any thing, is an undeniable argument for its possibility, and is alone a refutation of any pretended demonstration against it.".

Why does cursor use 10 tool calls in places where it would have used one call several months ago?.

I could go home but can I pass on this deal?

Why did cursor start using a lot more individual tool calls to edit single files? Like compared to a couple a months ago the change is quite noticeable.

In case you didn't know.

Google docs ai features are so much worse than what cursor offers devs. It is literally easier to edit a professional document in cursor than it is in Google docs.

O3 for analyzing images,.UI is great and it works.

You read the Temporal Differences paper by Sutton and then you read the trpo paper by schulman and the contrast is so stark. One paper is optimized for understanding and one is optimized for making the authors look very smart. Maybe it’s the general trend over those 30 years.

What does 10m context length even mean? You can say the context length is whatever you want. Transformers support any context length. When they say 10m context length they definitely don’t mean that all their training samples were 10m tokens long.

The way cursor doesn’t properly grab notebook context is beyond frustrating.

An llm Diffusion model that predicts layer by layer trained on activations of sota transformer.

The new cursor UI + sonnet 3.7 is fun.

Very annoying that deep research does not properly display mathematical terms. @ChatGPTapp.

PSA from deep research on “residual terminology” (how did residual connections come to refer to the skip connection).