That’s pretty cool indeed but intuitively I would have though “the video will include me clicking somewhere, and possibly having frames of stuff that’s not part of the problem” - so if I had a similar idea, I would have pasted 12 copy/pasted text fragments, or done 12 screenshots.
I wonder if this would have worked just as well (probably), taken more time (probably, but not much), and only scaling this up to 50 emails the video would have won out by a lot.
This is cool, and it kinda makes me think about possible data migration systems: say I want to move from Saas A to Saas B, but SaaS A has no API that I can get data from. SaaS B could implement a video scraper to ETL my data out of the other platform. I guess they could implement a traditional scraper, too, but that might be a lot more work.
This is probably a thing already - I just hadn’t thought about it
@simonw did you try reducing the screen time on the emails with the desired data? If so, how were the scraping results?
I am interested on that particular question in order to know how careful one should be with the screen capture and to even further reduce the Gemini costs.
I waited about 2 seconds on each email because I know that when Gemini first launched in February it treated videos as one frame per second sequence of images.
I think they may have changed that now but I wasn’t sure if they had.
That’s pretty cool indeed but intuitively I would have though “the video will include me clicking somewhere, and possibly having frames of stuff that’s not part of the problem” - so if I had a similar idea, I would have pasted 12 copy/pasted text fragments, or done 12 screenshots.
I wonder if this would have worked just as well (probably), taken more time (probably, but not much), and only scaling this up to 50 emails the video would have won out by a lot.
This is cool, and it kinda makes me think about possible data migration systems: say I want to move from Saas A to Saas B, but SaaS A has no API that I can get data from. SaaS B could implement a video scraper to ETL my data out of the other platform. I guess they could implement a traditional scraper, too, but that might be a lot more work.
This is probably a thing already - I just hadn’t thought about it
@simonw did you try reducing the screen time on the emails with the desired data? If so, how were the scraping results? I am interested on that particular question in order to know how careful one should be with the screen capture and to even further reduce the Gemini costs.
I waited about 2 seconds on each email because I know that when Gemini first launched in February it treated videos as one frame per second sequence of images.
I think they may have changed that now but I wasn’t sure if they had.