Multi reposted this
"Designing Usefully Unintelligent AI Summaries" We nearly shipped AI chat as our 1st LLM feature in Multi. Then we got over ourselves and shipped AI summaries. Was that the right call? Each day I become less sure. So I'm sharing our design thinking blog post now, before I disagree with all of it 🤣 We started with the observation that many AI features fail because they're too open ended. - They require too much "intelligence" from both LLMs and users. - Users have to manually provide all the context. What if instead, we could build an assistant that - Understands context of what you're trying to do - Can take actions for you - Is proactive, so you don't have to think about what to prompt If you're interested, check out the blog post. I'd inline it here, but I there are tons of useful images in there that LinkedIn won't quite treat right. https://lnkd.in/g6Eiz3VH Let me know what you think!
Thanks for sharing the depth here. Relevant! cc Wojciech Zalot Helder Araújo Alim Maasoglu
Technical Founder (LLMs, Rust, TypeScript, 10+ years in web)
7moThis is a challenge we're overcoming with AI at Forethink as well. I've taken the thinking that LLMs should present the context and simultaneously present ideas for the path to take to resume working in that context.