ChatGPT

• Opting out of Open AI using your data: You should opt out of Open AI using your data to train their model before using Chat GPT at Cognito. Please ensure you fill out this form to Opt out using the email address which you use to login to Chat GPT.
• Private and confidential information: Chat GPT 3.5 and later versions of Chat GPT can still use any data you input to train their model. Please ensure that you only ever input public information into Chat GPT, even if you have filled out the opt out information form above. 
• Inform the client: It’s important to grasp an understanding of how the client feels about Cognito using AI on a project - normally your line manager will be able to advise. We welcome use of AI for first drafts, not final drafts, but it is important that clients are comfortable and confident with our use of AI. If they have an internal policy against their employees using AI we recommend you avoid using AI for that specific client. 
• Old data: Chat GPT trained their model on data running up to September 2021. Thus, it has not been trained on any information from that date forward. If you are working on a project that relies on new policy updates, or information specific to events after September 2021, please note that Chat GPT has not been trained on any relevant information. 
• Inaccurate information: Chat GPT was trained on a select group of sources including Wikipedia, Reddit, Scientific Journals and more. It can be relied on more generally for general knowledge queries, but it can easily give misleading answers within our space and about our clients or their competitors. It can also produce inaccurate or false information as though it were true - please sense check any output from Chat GPT. 
• Biases: Several sources were used to train ChatGPT, some of which have obvious biases. These biases result in some outputs (e.g., racial or gender stereotypes) produced by the tool. Make sure any work you produce that has been informed by ChatGPT is not biassed.
• Copyright Infringement: ChatGPT uses a number of copyright-protected sources. Therefore, ChatGPT may produce outputs that contain copyrighted content. In addition to being an ethical issue, this could also pose a legal issue. It is difficult to know when copyright has been infringed with ChatGPT since it does not provide citations for the sources it was trained on. Therefore, you should use ChatGPT's output to inform your content but not as a final draft and always run your work through a copyright detector before sending it to your line manager or client. 

Sources: There is a lack of transparency about the sources the tool was trained on and about its decision-making processes (i.e., it is unclear how it comes up with certain outputs). Therefore, you always need to sense-check the sources by searching the information in Google afterwards and verify all the information it provides with credible sources.