I had the simple assumption that since it was trained to output one word at a time that was all it could do. That it was good at looking far back, but never ahead.

Will be interesting to read up on their evidence for this claim.

www.anthropic.com/research/tra...

Here are two solutions to advent of code day 1 2024.

Inefficient way of counting in part 2. But goal is minimal yet readable to make the reddit crowd happy.

And there is something pleasing with a conceptually minimal solution. Not code golf. But no fluff.