Chapter
Critique of Mechanical Turk Interface
40:35 - 45:22 (04:46)
The Mechanical Turk interface is poorly designed, with an excess of clicking and a lack of intuitive segmentation, which makes the experience of using it tedious and frustrating.
Clips
The success of Automated Speech Recognition (ASR) greatly depends on having large amounts of high-quality annotated data.
40:35 - 41:30 (00:55)
Summary
The success of Automated Speech Recognition (ASR) greatly depends on having large amounts of high-quality annotated data. Rev, whose business model involves annotating data, aims to achieve a 3% error rate and is confident in their ability to do so with their access to the best data for training ASR models.
ChapterCritique of Mechanical Turk Interface
Episode#151 – Dan Kokotov: Speech Recognition with AI and Humans
PodcastLex Fridman Podcast
The use of recurrent neural nets in drawing polygons has been explored in a few experimental papers, attempting to learn from human clicking and fixing of polygons.
41:30 - 43:07 (01:36)
Summary
The use of recurrent neural nets in drawing polygons has been explored in a few experimental papers, attempting to learn from human clicking and fixing of polygons.
ChapterCritique of Mechanical Turk Interface
Episode#151 – Dan Kokotov: Speech Recognition with AI and Humans
PodcastLex Fridman Podcast
The Mechanical Turk interface is difficult to use and AWS has done a poor job with it.
43:07 - 45:22 (02:14)
Summary
The Mechanical Turk interface is difficult to use and AWS has done a poor job with it. However, despite this, the future of online workforce platforms is exciting.