Otter AI Chat acts like a collaborative ChatGPT tailored for meetings

AI-transcription company Otter.ai has announced the launch of Otter AI Chat, a new AI-powered meeting tool that can provide real-time, question-and-answer support and generate meeting-specific content.

AI Chat acts as a meeting participant — answering questions, collaborating with attendees, and generating content based on meeting data. Users can message each other and the AI tool via the new feature, turning Otter AI Chat into an active participant during meeting discussions.

According to a recent study undertaken by Otter.AI, over 70% of professionals are already using some form of AI in their work and over 86% believe their work will look much different within a year because AI tools will revolutionize how teams work and collaborate. The study surveyed 1,500 US and UK-based office workers across a variety of industries.

Most of the AI technology that has been making headlines recently come in the form of Large Language Models (LLMs), such as ChatGPT, which are trained on public data or the internet. This means they don’t have a company’s internal knowledge, said Sam Liang, co-founder and CEO of Otter.ai. LLMs provide an interaction that is just one person talking to one AI model, a dynamic that is very different to the reality we see in most meetings, where  multiple people talk to each other.

“This is why we have designed a new chat model that supports multispeaker AI chat, making AI a participant and getting it involved in the conversation,” he said.

Otter AI Chat has three key functions: to provide answers, to facilitate collaboration, and to generate content.

While AI-powered summarization capabilities have been launched since the start of the year by almost every other company developing productivity applications, Otter’s new feature differs in the fact it can summarize and answer questions about what is being said in meetings in real time, as opposed to providing an overview of written documents or a summary of a meeting after it occurs, according to the company.

During meetings where Otter AI Chat is a participant, attendees will be able to ask the tool about discussion  points or key decisions made previously during the meeting — for example, whether there were any questions asked about a specific project or event. Using context available from the automated transcript, Otter AI Chat will be able to provide answers.

Additionally, during a meeting, attendees can communicate in writing to each other and with Otter AI Chat to seek clarification on particular points without interrupting the meeting to ask questions out loud.

After meetings are over, Otter AI Chat can generate action items, summaries, follow-up emails, blog posts, and other content that participants might need after the meeting has concluded.

“Otter AI Chat has knowledge about the user’s own meeting data meaning it can provide a much more personalised service to users,” Liang said.

Creating Large Spoken Language Models

According to Otter, the platform uses its AI to transcribe over 1 million spoken words every minute, providing a data source for the development of what Liang says the company is calling Large Spoken Language Models.

This model is based on millions of hours of conversational data, Liang explained, adding that this makes the model more difficult to develop as conversational data has a lot of differences compared to the written documents traditional LLMs are trained on.

“Written documents are usually more formal, they’re very systematic and have a certain structure,” he said. “But verbal communication, that tends to be more dynamic and involves multiple speakers… intonation and sentiment makes a lot of difference, as even the same word said in a different way can mean different things.”

Liang said that while Otter has adopted the term Large Spoken Language Model, it that doesn’t mean this type of model will be used exclusively by the company, but that Otter is able to move fast in developing the technology because it already has a massive data set of transcribed meetings readily available for training purposes.

“We have a pretty strong advantage,” he said.

The transcription and summarization capabilities for Otter AI Chat have all been developed in house and the company is using a combination of its own systems and some external APIs for the building of its Large Spoken Language Model, although Liang said confidentiality agreements mean he can’t divulge specifically who those third-party vendors are.

He also stressed that from day one, Otter has had strict security and privacy policies that ensure the user owns everything and that data can only be used for training purposes if the user specifically opts in. Additionally, any data processed through Otter AI Chat is only stored in Otter and isn’t available for use by any of the third-party vendors whose APIs have been deployed.

Otter AI Chat will start rolling out to users from today, with the expectation that every user will gain access to the functionality within the seven proceeding days. Otter AI Chat will be available to all users across all plans however, free users will be limited to five questions per meeting. Paid users will have no limits.