I watch the new course “Generative AI for Everyone”. I like the content, even if it is somewhat high-level. Until now, there is no special course-related board for this course, so I use the general board.
I do mind the many mistakes in the machine-generated English text transcripts, that are also in the subtitles. They are annoying to readers and foreign language learners.
This will also hinder the planned translation efforts as the text transcript (and not spoken audio) will be the natural basis for translation. It would be a pity, if errors and glitches in the English transcripts would spread to the translations.
What is a constructive way to deal with that? Is there any function to correct these mistakes in the English text transcripts as a user right on the spot and submit them for review?
Hi Ralf, and welcome to the community! You can find the dedicated category for the course under the Introductory section on the left sidebar. It was just added so you can navigate to it easily.
Thanks for reporting the subtitles. Those are generated by Coursera and we do realize there are some mistakes. We can assist in fixing them. If you spot any error, you can use the Report an Issue option at the bottom of the lecture videos. Select Subtitle Issues then paste the problematic phrase (and optionally the corrected one). That will be fixed after review. Thank you!
Thank you for your reply, Chris, and your recommendation.
However, pasting each problematic phrase and a correction is not going to work, as there are about 50-80 of them per video: issues with either single words or punctuation, marking direct speech, proper nouns, abbreviations, technical terms.
So, I just wrote a corrected version of the whole English transcript for Week 2, Video 3 “Lifecycle of a generative AI project” and submitted the whole new transcript. I used the issue reporting feature on subtitles, as you recommended. This shall serve as a test, whether this is acceptable for you. If so, I will likely send more transcripts to other GenAI4E videos.
Please let me know, if this method works well with your tools and workflow.
I like the light way it is being presented, but having taught more than 1000 executives in my life on AI, I just want to share some points from the course that can be better taught.
X framework is weird, learning strategy versus generative AI. gen ai is mostly unsupervised.
X LLM are non-supervised, this is a wrong interpretation of what supervised models do.
X LLM are not the tools to find information. It finds patterns on sentences, and can not be trusted for retrieval. Unless you put it in the context, to force st staying close to the truth that is pre-selected.
Hello Martin, I intended this thread not to be on such matters that you bring up that regard the actual content and statements. I think, your comments are valuable, somewhat debatable however (there is also a concept of “semisupervised” learning for example) and these matters should be in a different thread and you should say, what video lectures you refer to. Then your issues can be discussed separately.
Instead, this thread is intended to be on the pure transition from audio (spoken words by Andrew Ng) to the text transcript and not challenging the content itself.
My aim is, that instead of an AI-generated wall of words, the transcript should have: proper words (such as “in depth” instead of the AI understanding “in death” or “you.com” instead of “startup.com” or other misdetected or missing words), proper punctuation (meaningful sentences, commas separating subclauses to break up long sentences), proper direct speed (such as refering to example sentences, which should then be quoted) or marking proper names in quotes ,so that (automated) translators have a clue, that this is a proper name, instead of trying to translate names such as “you.com”. Also dealing with abbreviations such as “RAG”.
Hello, I’m from Brazil, my English isn’t fluent, so I’m downloading the videos, translating the subtitles and incorporating the subtitles in Portuguese again. It’s working very well.
Hi Ralf. Sorry for the late update here. It seems Coursera wants to handle the fixes so we’ve put off corrections in the meantime. Nonetheless, they saw your flag with the transcript and might include it in their workflow. Hopefully, we’ll have more clarity soon. Thanks!