I found the part about the voice and image stuff not being amazing weird - I thought they still haven’t released the new voice multimodal model to the apps. It

I found the part about the voice and image stuff not being amazing weird - I thought they still haven’t released the new voice multimodal model to the apps. It seems to lose the point of the change going from a pipeline task delegated to a VTT model to the LLM to a TTV model into a new voice / text multi modal model that can encode nuance in the audio into semantic space and impact generation of tokens which themselves being in the voice domain directly can encode generative nuances in voice. This can include picking up on and generating inflection in the voice etc. Maybe it doesn’t address software engineering by AI, but IMO, that’s the absolutely least interesting use case for AI on earth despite being manically obsessed over by software engineers.