ChatGPT plugin, suggestion

Can Anki create a chat GPT plug in? This way, we can upload documents to Chat GPT, which then creates flashcards that are integrated into Anki for more effective learning. I know that AI can handle processing multiple variables better than humans and will thus be able to utilize Effective learning: Twenty rules of formulating knowledge. I believe this has great potential beyond what can think for learning.

Thank you,
MA

If you ask chat gpt to output flashcards (e.g. question & answer pairs) in the TSV format, then you can import them back into Anki without needing a plugin. So it’s probably more a problem of coming up with the right prompt than integrating the two apps.

2 Likes

I recently created chatgpt plugin, will soon release it!

1 Like

honestly, from my experience, this is far from ideal.
some Qs are stupid, some are incomplete.

there is already one on youtube (seems from Brazil) wrote such a python and turned into .exe,
which could do this. this is using the chatgpt v3.5, not 4.

2 Likes

To people who were interested @krstoevan @tatsumoto @marafeh0 : I did it! Please be my first users!

https://ankiweb.net/shared/info/1416178071


4 Likes

someone from Brazil also made one, this use API key, so most people could only use GPT3.
(GPT4 api key need waiting list)

I tried,

  1. people say chatgpt (web version) is smarter than api key version
  2. web verison could be gpt4 if you paid, api key version usually is still GPT3 as stated above.

a gpt4 generated summary is here: The video titled “GPT Anki Helper Update” is an update on the development of a desktop app. The speaker provides a link to download the app from a GitHub page and explains that there are different versions for Mac and Windows. A new feature that allows resizing of the app window is mentioned.

The speaker discusses the use of a lupus diagnostic criteria system and the need for improvements in managing unstructured information. They suggest using GPT to generate a question bank based on unorganized information. The video also mentions clinical findings, joint manifestations, skin and mucous membrane manifestations, and cardio-pulmonary manifestations of lupus.

The video discusses the limitations of GPT and why the author has not implemented a certain method of generating Anki cards from a PDF. The author explains that GPT does not know what is good or bad, real or not, so it is not reliable for determining what is important in generating Anki cards. The author suggests that the best person to determine what is important is the user themselves.

The video was published on May 9, 2023, and as of the time of the summary, it has 136 views, 4 likes, and 8 comments. The video duration is approximately 9 minutes and 35 seconds.

anyway any share is appreciated,

i also got some scripts/macros made available on github, mostly on mindmanager.

oh, ok, i thought the OP asked for “paste text and ask for anki”.
it seems yours not exactly this, yours is more on language.

the one i refering to is for med students to paste text and ask chatgpt to make anki cards.

anyway, the result, at least using api key mostly use gpt3, is not very good.

Thanks for the link, i will add it to the list of alternatives. This one looks interesting, it seems that it targets creation of the new cards whereas mine is exclusively about filling the existing fields.

Yeah, for now I’m limited to use GPT-3 model. As soon as they allow, I will switch to the 4th model. I thought of also adding options that could allow the power user to pick the model and the options, or even the fine-tuning.

remind: sometimes the results are not satisfactory, so i myself prefer asking in the web version i.e. chatgpt, so that if anything goes wrong, i can immediately ask gpt to re-do according to the problem. this is the most optimal one.

Hey everyone,

Having read through the discussions here, I wanted to offer my perspective. While the efforts behind the Anki add-ons are commendable, I genuinely believe a dedicated ChatGPT plugin might be the way forward. Here’s why:

  1. Cost Efficiency: One of the main challenges with the current add-ons is the recurring cost associated with each card request on the OpenAI API. A dedicated plugin could potentially streamline this, making the process free for the users.
  2. Interactivity: ChatGPT’s strength lies in its interactive nature. Imagine a plugin where you can directly converse with your deck, getting real-time explanations for each card without the need for manual input.
  3. Seamless Integration: With a ChatGPT plugin, adding cards could become a breeze. Instead of manually creating cards, you could instruct ChatGPT to generate them, complete with its own detailed explanations. This would not only save time but also ensure a consistent quality of cards.

The current version of the add-ons feels like we’re adapting ChatGPT to fit Anki. In my view, it should be the opposite. With the rise of AI, we’re witnessing a paradigm shift. We should be allowing ChatGPT to make any changes we request in our deck without limitations, capitalizing on the interactive nature of AI. And to address potential concerns, implementing a backup feature that reverts cards to their last secure save would be crucial. This way, even if ChatGPT makes an error, we can quickly rectify it.

The only problem is that for this, we need the ANKI team, as we don’t have full access to the backend.

Not really though. Anki is open-source and you can access anything you could ever possibly need to access to play with your cards as much as you want.
What limitation do you see that requires the Anki Teams attention?

2 Likes

Well, for now the workflow looks like this:

  • you create a backup of your anki database
  • chatgpt add on fills a single field in a set of notes
  • in case you messed up, and rewrote a field that you should not have rewritten, you load from the backup

Right now the Ctrl+Z feature is not implemented and frankly I give it relatively small priority. If I see more people feel the need in Undo feature, I may implement it.

As for automatic creation of the new cards, there’s another plugin that does just that.

You propose tighter integration with anki, could you elaborate on how you imagine it?

If anyone is interested to try, I created a chatbot on telegram that uses chatgpt to verify the answers to your flashcard. It can also create flashcards based from a body of text.
You can try it out by searching for gumino_mem_bot in telegram

Hi! I’m the developer of AnkiBrain, which is to my knowledge the most integrated chatgpt plugin/addon for Anki. The integration is complete and there is no external tooling, manual work or fiddling to operate it once installed - for example you can send AI-generated cards directly from AnkiBrain’s interface to any Anki deck with custom tags.

There is a local mode (use your own OpenAI API key, ends up being very cheap) or server mode (requires a balance to make requests; I give free credits to users regularly). Unfortunately can’t afford to subsidize 100% free usage for everyone 24/7 but doing my best to make it as accessible as possible!

not sure you are talking api key for gpt3 or 4.

and i am just wanna discuss on the price of api key.

i use chatgpt plus plan, which now give 8x50x30 prompts, each with 2k token?

if this amount of prompt was used with api for gpt4, it cost over 10000USD, but chatgpt plus plan only charge USD20/month.

and i must say, gpt4 is much much smarter than gpt3.

AnkiBrain can use either GPT 3.5 Turbo or GPT 4. If you are using your own API key then it depends on your access. If you are using server mode you can use either.

AnkiBrain now has over 1,000 users and many users have told me that GPT 3.5 Turbo has been great for their needs. This is also been the case with my personal usage of AnkiBrain to learn advanced medical management (I am a senior resident physician at the moment). This is because AnkiBrain uses embedded document retrieval from a vector storage unit and supplies it to GPT with a default temperature of zero. GPT 3.5 Turbo is much more reliable in this case and rarely has trouble making cards automatically. I am still working on ways to improve it further of course.

If you are using GPT 3.5 Turbo via API key, the costs are so small that you will not end up accruing a large bill unless you are trying to make tens of thousands of cards every day out of every textbook you find. GPT-4 is definitely expensive but worth it specifically for programming, but not as much for medicine in many cases - I am happily able to gauge both with confidence given medical career and computer science background.

I doubt most users of the online interface plus plan are using it to the absolute maximum limit 24/7. There’s a good reason why they have closed down all unofficial APIs that backdoor into using the web client for chat completion and require usage of the official API. Also OpenAI is reportedly operating at a big loss anyway so a lot of this is getting subsidized by them.

I’ve had a similar experience, especially if you ask it to make cloze cards, it’s not really sure what to occlude. It is REALLY good at summarizing though, I’ve found that if I have a bit of lecture text that has a lot of fluff it makes it a little easier to think of a good card to make.

$540 million loss, yikes… seems weird though, if there are 1.5 billion active users and we suppose that maybe 1% of them are plus subscribers, they ought to be raking in $300 million every month. Something’s funky with these numbers.

yeah, other than the money issue,

i used to use python script and/or other people’s chatgpt plugin to summarize/make cards.

but sometimes those results in rubbish,

and then i found manually copy/paste them into the web chatgpt, i can direct it to where i want.

sometimes, i’ll add “timestamp 230816T204928” to the text i paste to let chatgpt know which text i am refering to.

ps: chatgpt plus’ now have code interpreter, it allow upload files, but recently it refuse to analyze the whole file at once, you need to paste the part you want to discuss to it, and it may reference to the uploaded document when necessary.

AI is still evolving.