We demonstrate that LLMs may learn indicators of document usefulness and modulate their updates accordingly. We introduce random strings (“tags”) as indicators of usefulness in a synthetic fine-tuning dataset. Fine-tuning on this dataset leads to implicit meta-learning (IML): in further fine-tuning, the model updates to make more use of text that is tagged as useful. We perform a thorough empirical investigation of this phenomenon, finding (among other things) that: (i) it occurs in both pretrained and unpretrained LLMs, as well as on a vision task; (ii) the effect of IML is small but significant; (iii) larger models and smaller batch sizes tend to give more IML. We explain and demonstrate how this IML effect may be attributed to the recently uncovered implicit gradient alignment effect of stochastic gradient descent-based optimizers. We also use probing to examine how IML changes the way models store knowledge in their parameters. Finally, we reflect on what our results might imply about capabilities, risks, and controllability of future AI systems.