• This thread is just the tip of the iceberg.The people ahead of the curve aren't Googling for answers — they're already in here, having the conversations you haven't found yet. DealerRefresh is free.Get the full picture →

AI = Awesome Intelligence

Am I allowed to say that I am developing a realtime realistic video avatar that can dynamically change the UI to answer user queries and gather details? The loom link is here.

Point is, typing is a hassle with chatbots and when a user calls he expects a human not an AI. As users get savvy about AI calls, its gonna have a blowback. But when they willingly interact with an AI avatar which helps answer their questions and even fill forms they don't feel cheated by it. Its the most ethical way of using AI and then handing off warm lead to a human. Not only that because it is visual can show relevant content by controlling UI its arguably better UX than a voice call. Not to mention calling is an extra step where large portion of website visitors fall off.

Curious to know what you guys think about it.
 
Claude power users, do you have a rich inventory of skills and plugins? Is your Opus/Fable output having bouts of obvious stupidity? Does it miss simple logical steps, does it invent steps?

Your skills and plugins maybe conflicting each other and/or the markdown content/structure was made for the older models when long, deep rules were needed to steer the output.

My output was so frustrating that I've spent 20+ hours getting it right and it's still not what I wanted. So, I created a brand new sterile Claude Max account and ran 2 identical deep analyses. Results:

"Your full Claude stack adds a strong interpretation layer. It also oversteers the model."

ugh. Back to the lab..
 

✨ AI Highlights

  • Claude and Gemini replaced ChatGPT for coding, planning, and research tasks
  • AI avatars with cloned voices fooled real people including spouses
  • Dealers built agents to stock inventory and find auction vehicles autonomously

Dealers and AI power users swap candid experiences with frontier LLMs like Claude and ChatGPT, noting a recurring pattern where model quality peaks at release then degrades until the next update. Key practical warnings emerge: 'yak shaving' (losing days building elaborate AI systems that turn out to be useless), plugin/skill conflicts that cause Claude to 'oversteer,' and the value of isolated testing environments to diagnose prompt stack problems. Tool recommendations and a real-time video avatar concept for dealership lead capture also surface.

Replies Views 63 14,931 Started Last Reply