Comment by UncleOxidant

> (the full impact of today's frontier models has yet to be felt by the general public - to say nothing of models that will be released in the next few years)

We definitely saw some kind of non-linear step function jump in quality around the beginning of the year - it's hard to express how good Claude opus/sonnet 4.6 is now. However, I wonder if we're going to see the same kind of improvement from here? It's kind of like we got to the 80% point but the next 20% is going to be a lot harder/take longer than that first 80% (pareto principle). Also, as more and more code out there is AI generated it's going to be like the snake eating it's own tail. Training models on AI generated code doesn't seem like it will lead to improvements.