Auto-transcribed captions needs improvement.

I suspect Adobe are aware of this, but adding my voice here in the hopes that others will too, since working with auto-transcribed closed captions is a pretty poor experience. Auto-generated captions often introduce so many errors, they barely save any time, and it's safer overall to manually caption videos to avoid difficult-to-spot mistakes.

• Poor and inconsistent orthography: Sentences often do not begin capitalised. Ends of sentences often have commas where there should be full stops, random commas appear mid-sentence etc. 
• Captions containing multiple speakers: Often, a single caption block will contain the words of two speakers as if a single sentence.
• Arbitrary/Inconsistent caption clip breaks: It's not unusual to have the first word of a sentence on its own, followed by two full lines of dialogue. Often, caption blocks extend well beyond the end of the dialogue. Gaps can appear between caption clips ever where there is no gap in dialogue.
• Arbitrary/Inconsistent line breaks: It's not clear how Premiere prioritises line breaks on multi-line captions. Ideally, lines would prioritise splitting by clause, followed word-count of lines. Function words such as conjunctions and preposisitons also would seem to serve as natural line breaks.
• Proximity to visual edits: Frequently, captions break within a few frames of a video edit, which can be jarring. It would be helpful if there were a feature to align caption breaks with edits if they fall, say, within 5 frames of each other.

It feels to me that this is one of the few tasks for which AI assistance would be enthusiastically welcomed by editors. LLMs surely could provide contextual understanding in order to significantly improve auto-generated captions.