LLM, but less harmful

This is my 7th blog post for the Blaugust 2026 community blogging project. Join me all this month for more blogging. Like every other asshole, I have an opinion about generative AI and LLMs. I think they can be helpful for certain tasks, but they raise significant, severe problems. This drives a lot of people to condemn LLMs stridently, shun them – and shun the people who use them. Others acknowledge the problems exist but wave them away, claiming that LLMs are too helpful to ignore. I’d like to stake a position between these extremes. The problems are real and I think we should fix them. I have some ideas. The problems with LLMs Let’s ackowledge right off that there are bad uses of LLMs. When LLMs are depended on to produce work, they can generate bad results and have terrible outcomes. It’s a problem when we displace human labor with machines that do the job worse. AI has negative impacts on learning. And those are just a start. But also, I claim there are good uses of LLMs. I’ll…