Forecasting potential misuses of language models for disinformation campaigns and how to reduce risk — PLINKFEED