memory-machines.com
Memory Machines: Can LLMs create lasting flashcards from readers' highlights?
We tested prompting, fine-tuning, RL, and grounded evaluation across ~1,500 labeled flashcards—and found models catch obvious misses but not plausible failures.