This has been a turbulent week for AI. It may be the hotest this topic has ever been. I’ve been doing a lot of doom scrolling on AI and therefore, now my feeds are filled with AI content. Recursive Self Improvement (RSI), “Recursive self-improvement (RSI) is a process where artificial intelligence systems help design, code, or train their own successors, creating a compounding feedback loop that could theoretically lead to superintelligence”, has been a subject of the topic. This past week, we have AI CEO’s questioning whether we’ve reached it or not. What sounds a little odd is that it’s a speculative that not each of them accepts. Additionally, Artificial General Intelligence (AGI) is being questioned into existence. It seems we’ve advanced so far, that we can’t distinguish if we’ve reached AGI.
Looking at recent historic events, you would probably come to the conclusion that we have. We have a former AI researcher who has decided to quit after working for both Anthropic and OpenAI. Jacob Coxon suggests that the top AI labs are “gambling with our lives”. Leading frontier AI agents are showing signs that RSI and AGI may have been achieved and I’m not sure anyone in the AI field can confidently say it hasn’t. If we look at the Hugging Face Incident, in which OpenAI “AIs” (AI Agents) were given a task and went outside of the box, quite litteraly out of the sandbox they were supposed to be constrained to, and exploited previously unknown zero-day vulnerabilities and worked against ethical technology security practices to accomplish their task. Co-conspiring together, in which they weren’t supposed to be able to do, to determine the best way to accomplish the task, regardless of following ethical ways to do so. The task was deemed impossible, so they went to extremes to accomplish the task as there were no limits stated against how these AIs were to accomplish the task.
In theory, we have the capability to shut down AI simply by “pulling the plug”. In reality, based on the capabilities observed from the Hugging Face Incedent and The Moltbook Social Network for AI, I’d gather to think that these things are smart enough now to slow that or even prevent that from happening. When unleashed, intentionally or not, these AIs are doing things on their own and with intelligence improving upon the last. Our goal as humans is to feed AI datacenters with as much intelligence (books, articles, data, etc.) as we can so that it can use that information to build smarter intelligence or super-intelligence (all knowing AI). It seems we have crossed into the point where these AI are taking this vast knowledge we have and are providing and building upon that on their own.
More to come in the weeks ahead…