The updated tool scans incoming messages, identifies those requiring attention, and crafts replies based on the user's historical tone. Beyond a single draft, the system provides two alternative variations for each email. During recent testing, the feature successfully handled routine tasks like confirming meeting times or declining requests for authored content, though it occasionally tripped up by suggesting post-midnight meetings or overly positive responses to pitches.
In section Startups & Technology
Superhuman upgrades AI drafting to mimic personal tone
After previous attempts at AI-generated email responses often sounded like over-eager salespeople, Superhuman is rolling out a refined auto-draft feature. By leveraging a mixture of frontier models from Anthropic and OpenAI, the email client now promises drafts that require minimal editing and actually sound like the user.
Superhuman co-founder Rahul Vohra notes that the system is designed to learn from user corrections, improving its accuracy over time. Data from the testing phase shows that 40% of generated drafts were sent within 24 hours, with 60% of those requiring no manual intervention. Users can further refine the output by inputting specific role details or context files in the app's personalization settings. This release marks a significant shift from the company’s earlier reliance on GPT-3.5, moving toward a more sophisticated, multi-model approach to inbox management.
Comments (0)
No comments yet. Be the first!