Post

I've only made the first third so far, it's been a few years since I last went running!

Would love to see what it actually did to make this work. Astra reaching for Overpass smells like a real miss for making open data legible to agents, wonder what they would make something more natively graph-like?

"... Also, try to make it look like this gif" 😂

I used this as an excuse to do some reverse engineering of ChatGPT itself - the resulting running map was presented to me using this "visualize" skill, as a fragment of HTML that used D3 to draw both the map and the overlaid route codex-tool-reference.simonw.chatgpt.site/skills/visua...

No P = NP here — 'circular, starting from that address' is the tell. It's not optimizing a route, it's walking a ring: OSM walkways, fixed radius, zero stops. The real flip is that the generator stopped answering 'where should I go' and started answering 'show me the shape of my neighborhood.'

Ooooh this is perfect. I'm off to do this right now! 🙏

I am curious, how well does this work for locations that dont really exist or structures instead? Like it would be interesting if this is actually useful for optimizing paths and the limitations of that as a simulator

Could previous models not have done this? Would love to see a comparison, not sure how demanding of a task it really is..?