It’s funny you say that. It was while doing a regular crossword that just for a millisecond I thought to myself, “Hey I bet Claude could answer that”, that I felt such disgust.
The crosswords are tough for a stock LLM to crack without invoking tool calls since they can hardly spell due to their tokenized inputs. But with tool calls, they’re getting pretty good at it.
It’s funny you say that. It was while doing a regular crossword that just for a millisecond I thought to myself, “Hey I bet Claude could answer that”, that I felt such disgust.
The crosswords are tough for a stock LLM to crack without invoking tool calls since they can hardly spell due to their tokenized inputs. But with tool calls, they’re getting pretty good at it.