Generative AI might look like a marvel but the reality is that behind its creation there is a dedicated workforce in companies at OpenAI and Google. These professionals are referred to as ‘prompt engineers’ and analysts who meticulously assess the accuracy of the chatbot responses to boost the performance of the AI.
However, the latest internal directive from Google to the contractors who are involved with the Gemini project has raised concerns about the potential for the system to disseminate misleading information on critical subjects, especially in healthcare to the general public.
To refine Gemini, contractors from GlobalLogic which is an outsourcing firm owned by HIitachi are frequently allotted the task of evaluating AI-generated answers based on criteria such as truthfulness.
Until recently, these contractors have the opinion to skip prompts that fell outside their areas of expertise which further allows them to avoid evaluating responses to highly specialized questions. For instance, a contractor without a science background could bypass a prompt that is related to cardiology.
GlobalLogic last week communicated a shift in policy from Policy and mentioned that the contractors can no longer skip prompts irrespective of their expertise level. Internal documents that were reviewed indicated that the previous guidelines allowed contractors to skip tasks if they lacked essential knowledge in such areas as maths or coding.
However, the new instructions make it mandatory for the contractors to engage with all prompts requiring specialized knowledge, advising them to rate the parts of the prompts they understand while noting their lack of expertise.
According to the new and updated guidelines, now contractors are permitted to skip prompts only in specific scenarios if they do not have complete information such as the full prompt of response, or if the content is harmful and requires special consent forms for its evaluation.
Thanks for choosing to leave a comment. Please keep in mind that all comments are moderated according to our comment Policy.