Video summary test 17: NHK 在野黨首論戰

This is a special program in NHK R1 (2024.10.11) after the new prime minister had his first speech. I tried to test the limit of chatgpt to summarize and extract how many times the topics had shifted.

https://chatgpt.com/share/670f103f-a6fc-8007-a2aa-4d60aa050ba6

The Japanese style of interviews or debate, is more like a replay for a rehearsal. All the topics and content are scheduled, there is no real key questions allowed for the speakers.

The mugshots just got more before election day


It also shifted topics quickly to avoid digging into deeper analysis and retro parts on what has been done. This hosting style focus on how to overwhelm first time listeners with an illusion that the things are going smoothly without blockers, with a wide coverage yet shallow mention for topics. It other words, to confuse and mislead participants with unnecessary details.


Here is a demo for it.I recorded it half way so you can imagine 50% more topic covered for such PR skills. You may download the full transcript here. 


At the moment of writing (2024 Oct), gen ai summary is not an ideal tool to dilute certain type of textual info - if we do not know how to ask sharp questions to the original transcription. The debate or transcription itself might not be worthy of summarizing at all when the speakers are dodging and forging facts. 

 

This is not about the hallucination AI has. Although it has been reduced, I still encountered several times. The AI we used are trained to hide critical info instead of digging out the deep insight we need from conversation. 

This is why agenda and tickets are important for other participants to tackle the hidden details ignored in such show. 



 

 

留言