[ExI] Do all AI models represent “cat” in the same way?

[Adrian Tymes]

On Sat, Jan 17, 2026 at 10:56 AM Mike Dougherty via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> https://pub.towardsai.net/toon-vs-json-a-comprehensive-performance-comparison-446a2fb82f20
>
> TOON is a data format that is 'cheaper' to communicate structured data to AI than the format programmers had been using (JSON) after it became obvious that it was better than the previous (XML) and archaic (CSV) formats.

JSON is optimized for robustness in certain ways that don't
necessarily apply in LLM token contexts.  For instance, in the latter,
you may be able to guarantee that newlines will only appear where you
want them, and they won't be inserted between tokens during
transmission; not all contexts JSON is used in can make that same
guarantee.  That's why JSON uses more characters to represent the same
thing.  (I also notice the comparison picture has at least 10 space
characters on the JSON side that don't need to be there, so it's not a
completely accurate comparison, though the general point is correct.)