## Sabotage Attempts in AI Training A group of experts in the field of artificial intelligence is exploring methods to "poison" the data used to train language models. The initiative, discussed on Reddit and reported by The Register, aims to compromise the integrity of future AI models. The idea behind this strategy is that by introducing incorrect or manipulated data into the training sets, the reliability and accuracy of the resulting models can be undermined. This type of sabotage could have significant consequences for companies and researchers who rely on these models for a wide range of applications. ## General context Training artificial intelligence models requires enormous amounts of data. The quality and integrity of this data are critical to ensuring that models are able to provide accurate and reliable results. Manipulation of training data is a serious threat to the responsible development of AI.