ChatGPT is taking us for a ride, showing us what is already possible and the astoning rate of progress of neural networks.
Most people are not aware of how good it already is and at what rate it is progressing. To understand, you have to query it, to ask questions to try to understand what it know and what it doesn’t.
So far, I find it very impressive. I asked it to write software in obscure details of a particular programming languages, I asked to write a poem, I asked to write a story, I asked to write a song. In most case it did way better than I could have done myself, at least in a reasonable amount of time.
What does it mean for the future of work? It mean that we all become managers of an infinite army of AI. We will have to manage them, to give them tasks, to give them feedback, to give them goals, to give them a purpose. We will have to manage them to make sure we are clear on what we want them to do and what we don’t want them to do.
One pitfall I saw many times is that the AI is a bit like a stranger you meet on the street. Even if he is a genius that know everything he can’t know exactly what you want with one sentence. As the AI doesn’t exactly know what is the level of detail or expertise you want it default to simple examples, simple songs, simple stories, kind of a human would do without knowing the context of use.
But once you provide it with a context, it can do much more. In my case I provided code of my program and asked to write new files to fix a gnarly problem I had. It did it right away, provided me a better solution that what I could find by googling and searching the relevant forum or asking for help.
There is so much low hanging fruit that can be done with AI, but it’s hard to know what will change and at what rate. I think an executive assistant could be extremely useful. A system that can read my email, order them by priority, know my goals and search the internet for me, prepare me the best information to help me achieve my goals. It could be a great to almost anyone.
It would be better if I could talk to it instead of writing, if it could understand me based on the tone of my voice, my posture, the expression of my face. That way it would be able to catch tiny cues that are not communicated in the text. It would be able to understand my mood and adapt to it. It would able to understand more what I want and when I want it.
That would make it ten time more useful. At the moment it’s like an assistan that it inhumanly fast and know everything, but that I can only talk to via the keyboard.
Another thing that is a bit annoying is that the system will always provide an anwser even if it’s not the right one instead of asking for more information or clarification. It always try it’s best, but don’t ask for more information.
It should have some level of confidence in how the response fit with what we want. That way we would teach him more accurately what we want instead of having to delete and start over.
On the vision front, I expect Tesla FSD to get to the level of a human driving and their robots to be able to do a few tasks in a factory.
This will put a lot of pressure on semiconductor fabrication and robotics fabrication. It mean that we can create usefull work as fast as we can produce motors, robots and the semiconductors to control them.
