Tiggered by Kaiming He's talk (www.youtube.com/watch?v=jEeL...).
Posts by Xiaotian Han
Moreover, as humans, we tend to evaluate others through the lens of our own personal experience. For example, a cautious driver might assume a stranger will drive similarly, extending the same level of trust they have in themselves.
Actually, AI can be regarded as one individual, but it is held to an unrealistic standard of 100%. Any accident involving AI completely destroys trust (trust is binary, not a percentage)
In contrast, AI lacks personal traits, so trust cannot be built in the same way. Instead, AI is assessed based on statistical measures (e.g., accident rates), which makes people more skeptical.
Humans are judged individually, while AI is evaluated statistically. A human driver’s presence implies they haven’t been involved in a fatal accident, leading people to focus on the individual’s perceived safety rather than overall accident statistics.
Thoughts on Why We Don’t Trust AI Like We Trust Humans (e.g., Autonomous Vehicles). Beyond commonly discussed factors like explainability, controllability, or experimental reliability, the often-overlooked issue is AI's lack of individualization.
It will be a special experience to present our paper at (tx.ag/gcnmixup) this Thanksgiving night at @logconference.bsky.social , as a TMLR Track Oral Presentation. Grateful to the organizers for hosting this at such a special time. Join us if you're available, and Happy Thanksgiving!
I'm hiring multiple fully-funded Ph.D. students for Fall 2025 and am interviewing candidates now.
If you're interested in Efficient LLMs, Understanding LLMs, LLM Post-training, or AI4Sci, please email CV and transcripts to xhan.hire@gmail.com. ahxt.github.io/group.html.
Thanks for sharing!
Hello Bluesky
Added Bluesky in my homepage 😀
would love to be added!!
would love to be added!!
would love to be added!!
would love to be added!!
would love to be added!!
would love to be added to part II!!
would love to be added!!