The company denied making “major changes,” but users report noticeable differences in the quality of their chatbot conversations.
The CAI subreddit since I’ve known it has been a loop of users complaining about the bot degrading, mods banning them, devs adding new redundant features, wave of toddlers flood in to replace the old users, toddlers stay long enough to notice the bot degrade, repeat
I notice this behavior in other AIs, JanitorAI (web) and PolyAI (Android/Apple), where they forget things spoken a few lines ago, or specific requests I make of a character. My biggest complaint is they forget their location or what they were doing, unless the author specifically scripted it. We’re getting there, slowly, but we’re a long ways away from Samantha (Her).
Yep. That’s why it’s important to understand how LLMs and other related technology work. Though to be honest, I’m not quite there either since I don’t have a computer science background. I just know that some LLMs can understand context more than others. You can check LLM benchmarks and customer reviews to see which LLMs fit your particular needs the most. For example, everyone is hyping over Claude 3.5 Sonnet.
As far as resembling Samantha goes, I agree that we’re very far away from that. In the movie, it’s acknowledged that she has developed her own consciouness and sentience. The same cannot be said about current iterations of AI chatbots. The more people, including AI companion users, understand the mechanism behind these things alongside with their limitations, the better.
