Modern "news" is a lot like sewage padded by corn fructose syrup and wrapped into shiny colorful pills to fool the mind. It's true that the pills have useful elements in trace amounts, but if you start harvesting them from the pills, you'll get poisoned. Your LM just removes the deceptive shell, but the pill's contents remain the same. I think it would be better if your LM identified and highlighted lies and deceptions in news articles. Most of them will be all in red, but that's the goal: give users an idea what they are reading. For example, the prompt for your LM could be: "for every statement in this arricle, mark it in red if it's a lie or an unsubstantiated claim, mark it in purple if it's appeal to emotions, mark it in green if it's appeal to authority" and so on. I doubt journalusts will be able to outmaneuver LLM trained on every book in the internet. (journalusts to distinguish from true journalists, that are exceedingly rare today)
though thats true this product (if widely used) would reduce incentives to make clickbaity headlines like 'You Wont Believe' or 'Watch what happens when'. Also 5000 word essay at the start of every boiled potato recipe. Thats a step forward but to change the actual content of news would require adjustment of journalistic motivations, which are primarily either money or power driven. they dont change easy especially given how much money is in keeping populus close to 50/50 on only 2 options.
