This is for people who mainly use chatbots and keep worrying they are already a step behind. You see 'An OpenAI model has disproved a central conjecture in discrete geometry' [C001], almost scroll past, then stop because maybe it changes what you should watch next.
The easy read is 'AI did harder math.' My read is different: the first thing AI is thinning is not jobs. It is field borders. [C002] The Erdos conjecture story matters because the move that cracked it did not stay inside geometry.
The write-ups used here point to ideas tied to Ellenberg-Venkatesh and Golod-Shafarevich. That is number theory, basically math about number patterns, not discrete geometry itself. In plain English, the breakthrough matters less as a flex and more as a sign that models can pull tools across subject lines.
One follow-up human paper pinned the improvement down as more than n^1.014. You do not need the formula. You need the sequence: AI opened the door by mixing fields, then humans turned it into a cleaner proof people can read and check.
Judge an AI update by whether it changes your next decision, not by how many features it adds. If you only use chatbots, watch for models that carry tools across fields. Scope: this take uses the OpenAI post and two linked paper summaries, not a full literature review or independent proof check. Share that signal, not just the headline.