Open-Assistant
Open-Assistant copied to clipboard
FAQ Optimization: Feedback Collection
Hello everyone,
I'd like to propose that we optimize and expand our Frequently Asked Questions (FAQ) to better serve our community. These FAQs are crucial in providing a foundation, guiding people in the right direction, and offering a quick overview of important information. By having comprehensive FAQs, it also reduces the workload of our moderators, as they don't have to repeatedly answer the same questions throughout the day. Instead, we can refer people to a central source of information.
To start this process, I would like to gather ideas for topics that you feel are important but currently not covered in our existing FAQs. Please share your thoughts about what is missing or unclear in this thread.
Currently, existing FAQ's:
- Discord FAQ Channel
- https://projects.laion.ai/Open-Assistant/docs/faq
I have one new question waiting in #1362, but two important questions might be "Does the Open Assistant model have any filters or restrictions built in?", as well as "Will the Open Assistant data set train my model to refuse to answer a particular type of request?"
Thank you, @horribleCodes - your support in this topic would be super helpful :)
To collect some topics/remarks I have in my mind and want to collelct here:
- Extraction of useful topics from https://docs.google.com/document/d/15veu2jXaEs90UsHUncKJBNlCDydSP7hv6Giqmw_ogQs/edit
- Linking to the FAQ from the main repo page on https://github.com/LAION-AI/Open-Assistant must be improved, here is currently a lot of good will need to spot the link at all
- You have to prove a similar good will to find the web FAQs at all. Docs are linked in the footer of https://open-assistant.io/dashboard, but you have to go through an intermediate landing page to get to the docs, where you can find the FAQ. We should consider linking directly in the footer of https://open-assistant.io/dashboard to the FAQs, 1 click instead of 3 would improve the user journey to this important information drastically.
- #1581 is also related to this, as it looks like https://projects.laion.ai/Open-Assistant/docs/faq isn't correctly shown on mobile
- How to interlink from the FAQ to the guidelines (goal: referencing the most important topics/entry points of it directly where this makes sense)
- Currently we have 2 FAQs - we must think about what parts of them should be in sync and what are platform dependent (should only been shown or discord or in the web browser)
- Maybe we should consider not only pointing out where bug/features should be captured, but also how (not overkill, but for people who don't quite know how to best do that)
- Perhaps we should also consider additionally using https://linktr.ee/ (or something similar) to provide an overview of let's say the top 10 most important links in the OA ecosystem - this link tree could be linked then from the FAQ too
- Again and again the question comes up how someone can contribute a dataset. This is described at https://github.com/LAION-AI/Open-Assistant/blob/main/openassistant/datasets/README.md, but you have to find it first.
- Based on this, perhaps it should also be said that help is gladly offered (and where) if someone feels overwhelmed by the instructions above when contributing a dataset.
- "How can I add a language" is a very frequently asked topic. We should not only link to the "how", but explain when this makes sense and how much data is needed for it to have any impact at all.
- Where people can find information about the current progress and will find in the future (web application) should be well linked, that is a topic which many people are interested in.
- How much training data is needed and what are the next steps/raw timeline in training the model based on the collected information is another very frequently asked question.
- In the license section of the FAQ, one could link to the existing discussions (issues) that have already taken place so that this topic is not repeated over and over again whether Apache 2 is the right license.
I think it is essential to enable people to help themselves. With good navigation aids (and also guidelines/tutorials etc.), I think it's possible, to reduce the barrier to doing this.
Very good ideas! I just wanted to point out that there's also a kind of short FAQ in the landing page (files are here).
@jentrialgo thanks for the motivational feedback :)
Additional topic for the collection:
- Many people are overlooking where the sign-in button is - including myself in the very beginning what I shamefully admit - "top right" on https://open-assistant.io/ could help many newly arriving new community members
A link to the FAQ really should be on the footer of the site - I think the only way to access it currently is through the doc, which could be bothersome and unintuitive to some people. Also, include a question about hardware requirements and methods to run it. We see question s about that subject nearly everyday.
Another possible question for FAQ:
- I see that many answers contain programming code without formatting. Are those types of answers correct?
@horribleCodes strongly agreed (FAQ link in the footer), because of this, it's already contained on the list I've posted above :) The hardware requirements are a nice addition I also saw the question countless times.
Some additional ones from my side:
- Even when people know that the initial version aims to run with 24 GB VRAM, they always get instantly worried because they don't have that much, while it was clearly stated that the model gets optimized in later stages to run also with less
- Message ID/Copying link to a message: People don't know that it's possible to directly link to a message and overlook this option in the dropdown when using it.
- What's the API key for the dev environment - this question was multiple times asked and seems to be unclear
Additional ones (first one inspired by https://github.com/LAION-AI/Open-Assistant/issues/1669#issuecomment-1434824407):
- How often does the leaderboard getting updated?
- When does a reset happening or are the intervals rolling?
Another one could be explaining the different states a message could be in - growing, waiting in prompt lottery - and how and when they move from one state to the other. I don't think it's explained anywhere in the docs.
Another question that I see a lot in Discord:
- Can I donate my CPU/GPU time for training the model, folding@home style?
The answer is that distributed training is not efficient, because it requires a lot of communication between hosts.
Thanks, @horribleCodes + @GuilleHoardings :)
Another one I've collected:
- I'm a developer, how can I support? - where it could be highlighted that there are good first issues existing
Another addition:
- Where can I find the changelog/How frequent are releases - not everybody knows https://github.com/LAION-AI/Open-Assistant/releases