Before pervasive GPS, it took me very little time to actually learn and internalize a route. Now it takes a lot longer to remember it when you're constantly guided. Same exact thing is happening with guided reasoning we get with LLMs
I have different experience. It took me some time to make a rote and write down all turns. Now getting from location A to B is a lot easier. Take a look at proposed rote and make some corrections. Meanwhile I spend time thinking about something else. So, GPS doesn't make me stupid or forgetful. It's just a tool which makes me more productive. The same almost true for LLM, except getting the right answer isn't always easy or possible. But overall on coding small utilities it's very helpful. For reasoning models I still need to find the right tasks. May be more complex utilities. Or the one I can't get from 4o yet: red-black tree with custom memory management and custom 'pointers' in data objects (small integers). While custom allocators are supported by std, the implementation still keeps native pointers, which locks it in memory.
I have different experience. It took me some time to make a rote and write down all turns. Now getting from location A to B is a lot easier. Take a look at proposed rote and make some corrections. Meanwhile I spend time thinking about something else. So, GPS doesn't make me stupid or forgetful. It's just a tool which makes me more productive. The same almost true for LLM, except getting the right answer isn't always easy or possible. But overall on coding small utilities it's very helpful. For reasoning models I still need to find the right tasks. May be more complex utilities. Or the one I can't get from 4o yet: red-black tree with custom memory management and custom 'pointers' in data objects (small integers). While custom allocators are supported by std, the implementation still keeps native pointers, which locks it in memory.