My interest in AI alignment keeps me tied to the AI space and how fast it’s moving. I’m not bullish on AGI being right around the corner, but I do firmly believe RSI is real and will lead to exponential growth. That part has been said a hundred times. What isn’t talked about as much is the change that even slight increases will cause in the everyday world.
I break those changes into two buckets: how current work will change, and what future work will look like.
Many jobs will almost cease to exist. There’s a lot online about which jobs will go, and some of it holds merit, but the main thing holding back systemic change isn’t the technology. It’s implementation.
I always draw a parallel to Japan. It’s a country with world-class hardware and infrastructure, but its corporate culture is still tied to tradition. Fax machines, hanko stamps, and lifetime employment are still the default at a lot of large firms. A lot of what they do isn’t optimal. Not because they’re stupid, but because they’re not forward-thinking about growth. Some fields will take longer to automate. Others will always be at the forefront of change.
I honestly can’t think of a single task in any field where using AI wouldn’t be helpful. Expert oversight to avoid hallucination will always be a thing. Nobody wants an LLM writing critical nuclear infrastructure code, though I do think Claude could have prevented the Therac-25 incident. So my take is that output per person will be higher, and oversight and taking responsibility will play a much bigger role than they do now, and the jobs where this isn’t possible will be the first to go.
Because of the research abilities of current frontier models, I don’t think you need deep knowledge in a field to do meaningful work in it anymore. Future work in almost every field will look a lot more like running a company. More responsibilities per person, more roles per person.
That’s as far as my understanding goes. I don’t think anyone can accurately predict where the boundaries of LLMs are, in terms of what they can do that humans can’t.