Right now it only works on texts that you uploaded which are up to 7,000 words long. It will run the whole text through ChatGPT and allows you to do things like fix spelling and grammar, fix formatting, or even simplify the language to make a complex text easier to read:
I’ve only just added this and haven’t tested it that much so please let me know if it works well for you or not.
If people like this and there’s sufficient demand, I may add support for longer documents. (To do this I’d need to split the text into chunks and process each chunk separately because the ChatGPT service has a limit on the amount of text that it can output in one request)
For now this is available to Premium users only, who can use it 5 times per day using the gpt-4o-mini model, and Premium Plus users who can use it 20 times per day using the full gpt-4o model.
Adding a space like “getuigdetegen” → “getuigde tegen” could be appropriate for formatting fixes I think.
Changing “op” → “in” and “dit” → “het” seem a bit more like spelling/grammar fixes its true. Are they actually correct though? If so, I’m not sure it’s a big problem. If I’m fixing formatting after using OCR say, then little fixes like this could plausibly be fixing mistakes that the OCR made. Adding the whole new word “houden” is a bit harder to defend.
Note that if you want to provide your own prompt you could deselect all the above options and insert your own instructions after checking “Other”. If you come up with a prompt which works well for you, please share it!
I’ve just made some changes which should make “Clean up formatting” less aggressive, particularly in the case where it’s the only option selected. Also, I’ve reverted the change so that Premium (non Plus) users will be using the smaller 4o-mini model again, for the full 4o model Premium Plus is needed.
AI isn’t totally predictable so it’s expected that there will be the occasional unwanted change, but please continue to leave feedback here and perhaps over time it can be made a bit more reliable.
Changing “op” → “in” and “dit” → “het” seem a bit more like spelling/grammar fixes its true. Are they actually correct though?
i don’t know, im not at that level just yet.
Note that if you want to provide your own prompt you could deselect all the above options and insert your own instructions after checking “Other”.
yep, i noticed it, but did not have enough courage to try it just yet
If I’m fixing formatting after using OCR say, then little fixes like this could plausibly be fixing mistakes that the OCR made.
clearly we have different perception of what formatting fix means
in my perception, formatting
fix is somethingwhen
you fix this kind of situation
and once you have your
formatting fixed
you just get a regular text with full sentences with a limit of your phone’s screen or whatsoever. what i normally ask deepseek to do (while chatgpt always almost tries to do smth what it wants) is this prompt – “remove the weird spacings/formatting without changing a single word:”
Adding a space like “getuigdetegen” → “getuigde tegen” could be appropriate for formatting fixes I think.
i think that already falls into “fix spelling/grammar” prompt.
i decided to try one more time, with the prompt “remove the weird spacings/formatting without changing a single word” and i also also added to make paragraphs’ separation more clear, i don’t remember what exactly i wrote there, the result was good, however, it corrected the typos (and only them) like the ones you see on the screenshot
what can I say now after a week once this feature was introduced? i love it!
now, whenever i import a transcript, i make the AI polish go through it just to make sure. it’s such a banger and a timesaver!
only once I experienced an error on my phone but i tried one more time and it went through well.
the only suggestion would be to have the “other” custom prompt saved (Reuse recent prompt) as it is in the reader.
(and another suggestion would be to have an option to ignore a specific change made by AI, but I don’t know how difficult it is to implement and at the same time it’s not that important, just sometimes it would be handy)
previously i used to use deepseek to do the same and one of the podcasters i mainly listen to talks quite frequently mentions countries of east asian region and whenever deepseek sees it it refuses to respond, even if the discussion goes in a complementary way. using chatgpt was not an option as it hallucinated constantly
One week cehck-in: You may want to tweak the prompt slightly as I have found some instances with YT videos where it would attempt to “improve” word choice in places where the transcript did not have structural issues. Despite minor issues, it is working well over all.
This is now possible with the latest update. If you don’t like specific changes that the AI made, you can click them to revert them at the review stage before clicking “Save Changes”.
Just tried it out for the first time and it worked great I used the “Clean up formatting” option on a Persian text that I got from Wikisource and whose formatting had been bothering me for a while.
As a detail obsessed control freak I compared both versions and all the proposed changes made sense to me. A lot was about punctuation (the original text was using a lot of spaces around punctuation marks) and occasionally more usage of zero-width space (the original text rather used spaces but contemporary Persian orthography prefers ZWSP for prefixes and suffixes, which is especially helpful for language learners).
I used the “Clean up formatting” option on a Persian text that I got from Wikisource and whose formatting had been bothering me for a while.