Its solving things that was unthinkably impossible to solve by AI.
Think of it logically. Before it could do 0% of our ability. With chatGPT it increases to 30%. With a domain specific LLM trained to target code... that increases to about 50%.
A person with 50% of their brain is obviously mentally deficient. It's not even comparable to human intelligence. It's like a stupid assistant. Helpful, but still stupid.
The issue here isn't really about the comparison between ai capabilities and human capabilities. It's about progress.
That 30% to 50% trendline in LLMs and other areas of improvement in AI is the Same trendline as we see in say creative writing, art, etc. Etc.
It points to the fact that coding is not inherently superior to other forms of creation because the trendline is exactly the same. In fact drawing really good art is actually harder then programming (no 6 months boot camps for art).
We are unfortunately on target for replacement because the next data point on that trendline is obviously not 51%. It's wishful thinking and illogical to ignore the obvious trendline.