Artificial intelligence powers Slack’s new search and summarization capabilities

Accessing institutional knowledge has proven difficult with traditional search methods, despite Slack’s status as an enterprise messaging platform and its de facto status as a repository for stored knowledge. A new AI-powered search tool and the ability to summarize content inside channels are two new capabilities that Slack unveiled today in an effort to increase the accessibility of that information.

Chief Product Officer Noah Weiss of Slack claims that company data is organically gathered by the platform in an unstructured and casual manner. Finding a means to reveal that knowledge treasure trove is the issue. “What’s really funny about all of this is that we can now extract a whole new set of intelligence and meaning from all of that analysis that has been created for years [on our platform] thanks to this new wave of generative AI capabilities,” Weiss told TechCrunch.

At the Salesforce World Tour in New York City in May of last year, Slack made the announcement that it was integrating generative AI into the platform. With the development of SlackGPT, a unique type of generative AI created especially for content on the Slack platform, it was more of a wide call to action.

The announcement made today aims to put that to use in more targeted ways. Weiss claims that having the ability to compress channel information makes it easier for workers to resume work after a break or just to skip reading a long thread in order to understand the main points of discussion. Weiss claims that an essential aspect of the feature’s design was the ability to request a summary of any topic discussed on Slack’s AI model, which is generated via channel summaries. The model’s creation process is documented with references, demonstrating how it arrived at each summary point.

Any area can be dug into, and we’ll show you each individual component of the context. Thus, we were focusing on establishing openness, fostering trust, showcasing our work, and enabling individuals to dig deeper into the information if they so desired,” he stated.

Like ChatGPT, the firm also lets users ask questions organically. However, it uses Slack content rather than more generic internet stuff, so a user may ask something like, “What is Project Gizmo?” After that, Slack AI provides a response, once more citing its sources so users may determine the accuracy of the information.

Every response has a quality check feature that allows users to rate the answer as good, bad, or neutral. This allows the system developers to review the model’s performance and the model to learn about the quality of the responses.

He would only remark that it was a combination of big language models, refusing to go into details about the underlying model. We discovered that each one operates differently, exhibiting varying speeds and quality features. We invested a great deal of effort in prompt engineering as well as optimizing the models for the data that we really have in Slack.

As an add-on product for enterprise plans, Slack AI with search and summarization will come at an extra cost on top of the standard license cost. Slack did not provide pricing information, but the business claims that it is currently available in the United States and the United Kingdom in English exclusively for the time being and will soon be available in more languages.