AI and programming classes in Llanelli

The learner downloads every mapped road and path around Llanelli from OpenStreetMap in one query and keeps the 559 named streets that lie wholly inside a rectangle drawn around the town. For each one the program answers a question from the map itself: does either end of the street stop at a point joined to no other road a car could use? By that test 215 streets, 38.5%, are dead ends. Footpaths are ignored, so a cul-de-sac with a path out of it still counts as a dead end for cars.

Next the streets are split by naming pattern. 293 follow the English pattern, with the kind of road as the last word: Road, Street, Close, Court. Only 71 of them, 24.2%, are dead ends, and the last word is a strong clue: 9 of the 11 Closes and 8 of the 9 Courts are dead ends, against 7 of 83 Streets. Another 122 follow the Welsh pattern, with the kind of road first: Heol, Clos, Llys, Maes. Here 72, or 59.0%, are dead ends, including 15 of the 23 streets starting with Clos and 11 of the 15 starting with Llys. The remaining 144 names fit neither pattern and are set aside.

On new English names the model looks useful, beating the lazy guess by 6.8 points. Its heaviest weights explain why: "close" pushes hard towards dead end and "street" hard away from it. On the Welsh names it is exactly as good as saying no every time, and in all 200 runs it does not find a single one of the 72 Welsh-pattern dead ends. Clos, the Welsh word for close, sits at the front of the name where a last-word model never looks, and every word after it is new to the model. Nothing about the streets changed; the clue disappeared. The map answers the question perfectly in a few lines of code, because the map, unlike the name, is the thing itself.

Llanelli teenagers who can test a model like this are ready to question AI rather than simply use it, and that confidence is built by writing code. The longer argument is in why code is still worth learning for teenagers in 2026.