3 ms·
Surely an ML pipeline could 1) listen to audio, 2) Identify the public official speaker from the tonality of their voice 3) transcribe it, 4) summarise it? W
by rorykoehler 8y ago
Surely an ML pipeline could 1) listen to audio, 2) Identify
the public official speaker from the tonality of their voice 3) transcribe it, 4) summarise it?
We may want a human in there to verify and edit but it would reduce costs significantly.
- thekyle 8y agoMachine summarization is still a very active research topic and is nowhere close to being robust enough to produce a full write up. The state of the art is able to write headlines correctly most of the time but not much more.
- perl4ever 8y agoMachine written articles are a large part of why news in general is virtually worthless to me these days. Especially in finance, I frequently read things that betray a total lack of understanding of the content. This has a knock-on effect that might not be obvious. As a journalist, you might think, well, why don't I read a reputable source like the WSJ - isn't good journalism more valuable when there is more bad stuff about? But people are getting clicks and selling ads with the automatically produced garbage, and therefore I know that traditionally trustworthy media has neither the incentive nor the morale to live up to their past standards. And thus, I have no interest any more in subscribing to even the best. For years, I've watched how under modern competitive conditions, a good brand (in any industry) is not worth maintaining any more, but rather it's become optimal to strip mine it to a husk. It's not really some mysterious decline of society, but a direct result of better data analysis. If someone says "we won Pulitzer prizes!" it doesn't soften me a bit - they only want me to pay them for what they did in the past, and that means nothing.