Block's Dessa AI: Raw Human Chat, Awkward Edges Included

·Autoro Tech
Block's Dessa AI: Raw Human Chat, Awkward Edges Included

Look, another AI model just dropped. Jack Dorsey's Block rolled out Dessa, and they're pushing it as something different. The pitch? It's trained "from scratch on nothing but human conversation." No pre-digested internet sludge, apparently. Just raw chat.

What does that mean in practice? Well, Block's been sharing some of Dessa's conversations with humans, and they aren't always polished. Ragavan Thurairatnam, Dessa's lead, put one out there that was frankly a bit uncanny, a little awkward. It's not the slick, predictable chatbot patter you're used to. That's the whole point, I guess. Friction point one: Expect some conversational speed bumps. It's not always going to flow perfectly.

For anyone building ad automation or customer interaction stacks, this is a curveball. Most of us are chasing efficiency, trying to smooth out every interaction. Dessa throws that out the window. It's designed to mimic messy human talk. That means your usual metrics for 'successful' AI interaction might need a rethink. Friction point two: You'll need new ways to measure engagement if the goal isn't just 'task completion' but 'realistic interaction.'

Think about the backend. Integrating something like this isn't just a plug-and-play. You're dealing with an AI that might throw unexpected responses. Your existing NLP pipelines could get tripped up. You'll need robust error handling, edge case planning, and probably more human oversight than you'd budget for a standard chatbot. Friction point three: The dev lift for integration and ongoing maintenance could be heavier than you anticipate, especially if you're trying to corral its "unconventional" style into a predictable marketing funnel.

Block's banking on this raw, unvarnished approach. They're betting that authenticity, even with its awkward pauses, resonates more than a perfectly sterile, predictable script. It's a gamble, but one that could shift how we think about AI in customer-facing roles. It's not about making AI sound flawless; it's about making it sound real. And real isn't always smooth.