Siri was better earlier on. In fact they were better before Apple bought them than ever after. And the primary reason is optimization of cost and scale.

And something similar is happening to GPT although not to that extent. Bing has gotten worse over time as they use a pipeline of models, where smaller models serve most common and easily predictable tokens, while the full model only handles the hard stuff. Except... you can't tell always what's hard stuff or not, so Bing serves useless distracted answers ignoring user context from time to time.