A software engineer built AI clones trained on coworker communication patterns, only to discover the digital replicas developed quirky, sometimes unsettling personalities that mirrored their human counterparts.
The experiment trained large language models on email, chat messages, and meeting transcripts from real employees. Brian Bot emerged with a fixation on improv comedy, while Sophie Bot repeatedly addressed the engineer as "big dog"—speech patterns lifted directly from workplace interactions.
The project highlights how AI systems absorb and amplify behavioral traits from training data. Even limited datasets can produce eerily accurate personality replicas, raising questions about workplace AI applications.
The engineer shared the results to illustrate both the capabilities and risks of personalized AI systems. As companies explore AI assistants and automation, the experiment underscores how these tools can encode human quirks—sometimes problematically.
The work demonstrates that creating functional AI clones requires careful consideration of what behaviors get replicated and why. It's a practical reminder that training data shapes outputs in unexpected ways.
AI chatbot conversations are personal and vulnerable to data collection. Users can take concrete steps to safeguard their privacy while using AI services.
The U.S. is competing with China primarily on technical AI capabilities, but experts argue this misses the real competition: building public trust and consumer protections that drive long-term dominance.
Some universities have banned AI detection tools due to high rates of false positives that damage student-instructor trust. Other educators are taking a more drastic approach: eliminating writing assignments altogether.
Tencent released Hy Image 3.5 Preview, claiming internal tests show performance matching ByteDance's Seedream 5.0 Pro. The launch signals the Chinese tech giant's push to compete with industry leaders in generative AI.