In June 2023, we introduced PollSense AI, Conferences i/o's first experimental feature leveraging large language model (LLM) tools, more commonly known as "AI".


We are excited about the potential of these tools to expand and improve the benefits that Conferences i/o delivers to customers, but we are moving slowly so that we can best consider latent risks, like the implications on data privacy, security, and intellectual property rights. We recognize that while AI has much to offer in terms of efficiency and productivity, it also introduces new questions about how data is used and protected.


To that end, we have prioritized transparent communication about our use of AI tools. In this document, we aim to clearly explain what's going on behind the scenes. Our aim is to ensure you feel informed and confident about using PollSense AI, knowing that the security and integrity of your data is a top priority.


TABLE OF CONTENTS


What underlying AI technology is Conferences i/o using?

Conferences i/o is using OpenAI's LLM models, with appropriate fine-tuning for specific needs.


Why did Conferences i/o select OpenAI over other options.

Our team began investigating AI tools in 2022, and has observed that OpenAI's products deliver consistent, high-quality output across a range of activities. For these reasons, as well as OpenAI's privacy-friendly policy shift in March 2023, we decided to move forward with building our initial AI implementation with OpenAI. That being said, we will continue to review alternatives as they become available, including in-house implementations of LLMs.


Was the AI model trained, or fine-tuned, using data from Conferences i/o customers?

No. OpenAI's models are pre-trained on massive datasets that would not include Conferences i/o customer data. As for fine-tuning, our team created polling content from internally-generated slide decks over the past decade. This fine-tuning guidance helps GPT-4 deliver more consistent responses.


Future AI implementations for Conferences i/o could potentially be trained, in part, using data from Conferences i/o customers. However, this would not be without customers providing explicit opt-in authorization.


What does Conferences i/o do with uploaded presentation data?

Upon receiving a PPTX slide deck, Conferences i/o extracts the raw content and integrates it into an API request to OpenAI. For quality assurance and auditing purposes, Conferences i/o logs both the request sent to OpenAI and the ensuing response. Importantly, these API logs, which include presentation and response data, are routinely purged every 30 days, ensuring no prolonged storage of customer presentation data.


What does OpenAI do with this data?

Data submitted via OpenAI's API (which Conferences i/o is using) will not become a part of OpenAI's training data.

Similar to Conferences i/o, OpenAI will retain log data on a request for up to 30 days.


Is OpenAI safe and trustworthy?

We think so. OpenAI has been transparent in their policies and maintains a thorough trust & security portal (https://trust.openai.com/). Further, third-party security audits suggest that OpenAI follows through in its commitments to security & privacy.


Does Conferences i/o or OpenAI have the right to use or share my presentation data?

No, at least not beyond the bounds of using presentation data to generate polling question.


Is there any way for PollSense AI to ignore particular slides within a presentation?

Not yet, but we are working to add this capability to Conferences i/o.


Is it possible to opt-out of AI tools in Conferences i/o?

As of this writing, AI tools in Conferences i/o are opt-in, so by default customers will not see them.


Note: If you have further questions about AI at Conferences i/o, please Contact Customer Support.