IntercomAnnouncement
Meet Fin: Intercom's AI chatbot, Powered by GPT-4
Summary
Intercom launched Fin, an AI chatbot powered by GPT-4 and fine-tuned for customer support, combining existing Resolution Bot technology with the new model. Fin answers questions solely using content from a business’s existing help center, reducing hallucinations by constraining its knowledge base and refusing out-of-domain queries. It requires minimal setup, automatically ingesting articles from Intercom or Zendesk help centers, and updates responses when articles change. Fin links to source articles for verification, handles multi-turn conversations, and seamlessly passes unanswered queries to human support representatives. Limitations include potential inaccuracies, high cost of GPT-4, and latency of 10 or more seconds, but Intercom expects improvements, is exploring configuration settings for creativity tradeoffs, and is rolling out a limited beta with a waitlist.
Context
Customer service teams needed a chatbot that could answer questions naturally without hallucinations, but earlier models like GPT-3.5 were too prone to making things up.
Approach / What changed
Fin combines Intercom’s Resolution Bot technology with GPT-4, constraining answers to help center content to reduce hallucinations, requiring minimal setup by ingesting existing articles, and linking to sources for verification.
Takeaways
Fin constrains its answers to content from the business’s help center, refusing out-of-domain questions to increase predictability and trustworthiness.
Fin requires minimal setup by automatically ingesting existing help center articles from Intercom or Zendesk, and updates responses when articles change.
Limitations include potential inaccuracies (though reduced with GPT-4), high cost of running GPT-4, and latency of 10+ seconds, but Intercom expects these to improve over time.
FAQ
How does Fin reduce hallucinations?
Fin reduces hallucinations by constraining its answers to only use content from the business's help center and refusing to answer out-of-domain questions. It also links to source articles for verification, allowing customers to validate responses.
What happens if Fin cannot answer a question?
If Fin cannot answer a question, it seamlessly passes the query to human support representatives, integrating with Intercom's existing workflows for a smooth handoff.
What are the main limitations of Fin?
The main limitations are potential inaccuracies (though reduced with GPT-4), high cost of running GPT-4, and latency of 10 or more seconds. Intercom expects these to improve over time and is exploring configuration settings for creativity tradeoffs.
Source and freshness
- Source
- intercom.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.