Time to Dig
Every other day seems to go like this: another frontier model drop, and the entire internet is in a state of psychosis again. The clouds of AGI are upon us yet again, and everything that we know about humanity is going to be automated tomorrow.
Before you say it, yes, I am well aware of the Millennium Prize problem(s) solved by closed models, the HF incident, and everything else happening at the frontier of this technology.
Having said that, I still believe the current growth spurt in AI models is going to leave all of us much smarter and more prosperous if we keep digging and using the models to their utmost potential. Based on whatever I keep seeing on the WWW, aside from the people working at the labs themselves, most of humanity is still debating whether we need this technology at all, or has limited the utility of the electricity of our age to outsource its work and thinking to the models, instead of using them to sharpen their craft. Here’s an argument for exploring this technology to its full potential, a lot different than the common doomer arguments that you keep seeing everywhere else.
What I do not understand is, when the when the bar is so incredibly fucking high, why would you not want to dig faster, harder, and better?
Why does everybody want to sell adjacent feature Y as an improvement to current workflow X, when X exists only because we had no way of doing it faster in the pre-AI era? Why does everyone want to be a plumber in this Renaissance of knowledge, when everyone can create works of art that previously existed only in our private or collective imagination?
The knowledge needed for projects that once seemed too hard to execute or required too much money, effort, and manpower, is now baked into the models, and starting anything new is just a click away. We literally live in the age of Bootstrap-ification (like the UI framework) of every single type of knowledge work that exists, and we have two ways to go from here:
- Keep replacing everything with AI till something sticks.
- Imagine and iterate on new paradigms that just weren’t possible earlier.
The problem with (1) is that it has been tested for millennia with every new piece of technology and has almost always failed.
One example is how factories were reimagined after electrification in the late 19th and early 20th centuries. At first, factory owners replaced central steam engines with central electric motors because electricity was the hip technology, producing no visible productivity gains for decades. Then they had the imagination and courage to add an electric motor to each machine, making every operator far freer in terms of location and task.
Another argument I don’t understand is this: if intelligence is the electricity of our age, why do we want to spend all our time debating whether it is good, rather than all the things we can now do that we didn’t think were possible? And honestly, if AI is destroying your critical thinking and long attention span, please show me the Nobel Prize that you received before the launch of ChatGPT.
If the frontier models have compressed all of humanity’s public knowledge, and now we can reach the baseline much faster, why are we not pushing the boundaries of possibility further?
Why do you want to sell the plumbing and wires in an age when we don’t even know what is possible in the future?
And if everyone starts selling shovels, then who will do the digging?