Info for Pilot Participants

Thank you for joining the Pilot for the CCV AI Transcribe service! We are excited about making this service available to the Brown community, so your feedback will be invaluable for us to improve this service!

Before you get started trying the service out, there are a few things that might be helpful to know:

Scope

The Pilot is for the Transcribe service only. You should also have access to our other AI services that we are developing as well, but these services are currently under development and are provided as-is. We do not guarantee that they will work, and they might change without any notice. That said, please feel free to check them out and feel free to let us know if you have any suggestions.

Using the transcription service

Please follow the Transcribe page for instruction on how to use the service. While you are using the service, please also feel free to be adversarial - that is, feel free to deviate from the instructions and try to break the service. We want the service to be stable to everyone once it launches, so if you have successfully broken it, we would like to know.

A few interesting things that might be helpful to try are:

• Unusually large audio files

• Files that we currently don't explicitly support (we only support .wav (PCM), .mp3, and .flac files at the moment, but other file formats might work)

• Audio files in languages other than English or Spanish

• Audio files with multiple speakers talking over each other

• Audio files with poor audio quality (such as recordings of phone calls)

Using the transcription results

You might already be wondering whether you can use the transcription results in your work. While the quality of the transcriptions provided by our models should not deviate too much from that of the final releases of the Transcribe service, we would recommend that you check the transcription results before using them directly in your work. Like other AI models such as ChatGPT, the models that we use can also be prone to hallucinations and mistakes, and with the open-source Whisper model, we do not apply any filters for potentially dangerous or offensive language, so please exercise plenty of caution and check the transcription results before using them in a different context.

About our carbon emission estimations

We are estimating the carbon emission of our AI models using the methodology described in the Sustainability section. We are still working on the carbon emission estimation for the Whisper models, and therefore, we do not have any figures for it yet. We will roll out this feature in the very near future.

Feedback and feature requests

Your feedback is paramount to the ultimate success of this project – perhaps to the CCV AI Services as a whole! During your time using the service, please provide anything of note you would like us to know. We would like to specifically mention the following feedback targets for us:

• Crashes of the service. Please include a time and date the issue occurred, and what steps you took to encounter said bug!

• User Interface issues or confusion.

• Completely unexpected transcription results (vs. a couple words mis-placed or lost).

Please submit a ticket of your issue or feedback to [e-mail] with your subject starting with "CCV AI Pilot:" so our fantastic Help Desk staff can direct this to the correct group.

Please Do Not Share

During our Pilot, we are dedicated to bug fixing and feature iteration. Because of this, we are trying to keep to a small but dedicated Pilot group; unauthorized users to the service would cause difficulty for us finding and fixing issues. Please don't share this Pilot with others - instead, send them the Intake Form so they can get on our list for access!