vok

The examples:

https://www.dangermouse.net/esoteric/piet/samples.html

The last one is astonishing: "Okay, this one is really wild. Piet J. (yes, that's his real name) was browsing art in a small gallery and saw a work which reminded him of a Piet program. He spoke to the artist, who claimed to know nothing about the language. Piet took a photo of the artwork (left), converted it into a clean image file using close colours from the Piet palette (right), and tried running it.

It ran!..."

Corrected link: https://www.youtube.com/watch?v=nO9aot9RgQc

Well, o3-mini-high just successfully found the root cause of a seg fault that o1 missed: mistakenly using _mm512_store_si512 for an unaligned store that should have been _mm512_storeu_si512.

I think that "Good Forth programmers arrange things so that they flow on the stack" has analogs in other languages. For example, arranging things in J so that short tacit expressions naturally provide the functions you need.

There's interesting discussion with the authors here: https://www.alphaxiv.org/abs/2408.10234

This "Computer use" demo:

https://www.youtube.com/watch?v=jqx18KgIzAE

shows Sonnet 3.5 using the Google web UI in an automated fashion. Do Google's terms really permit this? Will Google permit this when it is happening at scale?

With the right font (e.g. http://viznut.fi/unscii/) and line spacing, braille characters work fine with an even distribution of pixels and no gaps.

These strategies seem immediately practical. If you want to go beyond zero-shot for LLM coding, you may not need a complicated agent architecture - just start with escalation, retry, and warming.

APPS has 3 subsets by difficulty level: introductory, interview, and competition. It isn't clear which subset Claude 3 was benchmarked on. Even if it is just "introductory" it is still pretty good, but it would be good to know.