Reinforcement Learning is a weapon of mass destruction

Reinforcement learning is a weapon of mass destruction. Yet another idea in the research institutions for years is now rapidly being deployed in production and relevant to our everyday lives. While software developers are still arguing about the merits and uses of "AI", reinforcement learning has already been deployed in production in areas that can have massive impacts on society.<br><br>This doesn't matter for material sciences. It doesn't matter for medicine. It will matter vastly for economics, politics, warfare, and the potential destruction of humanity.<br><br>GPT-4 will be capable of "autonomous self improvement" - that is, it's capable of writing code that improves itself faster than humans can improve it. This just came out of "research" and is already being deployed in production. Google Gemini is just as good in some benchmarks as GPT-4.<br><br>Now, the recent news that OpenAI and Meta are both working on integrating Reinforcement Learning into LLMs means that they will be able to write the most viral, psychologically manipulative content on the internet. <br><br>And these companies are run by billionaires with no oversight. MarkZ is putting billions of dollars a year into developing Gemini. He doesn't care about the potential consequences. The CEO and CTO of OpenAI just left and they are blaming Microsoft for putting profit over safety.<br><br>Reinforcement Learning is a weapon of mass destruction because it's going to make it possible to write extremely manipulative copy. It's going to save corporations a lot of money in marketing and copy writing. It's going to squeeze out competition and smaller institutions. And the impacts are going to be felt this decade.