Maybe I don’t want a “safe/fair/unbiased” LLM
SAFETY
-
Neural Networks Need Kindness and Useful Content Transmission
By
–
You are both correct; the neurons need to be kind to each other (to keep everyone motivated to play their best game), while also rewarding and transmitting the most useful content
-
Generative AI Not Making Humans Obsolete Media Narrative
By
–
Humans are not being made obsolete by generative AI. This is a boogeyman that has been successfully created by parts of the media, and I suppose that in a well intentioned attempt to create awareness for X risk, the X risk community may have contributed to this
-
AI Guardrails Dilemma: Balancing Safety and User Reassurance
By
–
I know this is kind of goofy but it does illustrate how hard guardrails are when users can ask AI anything at all Do you reassure someone who seems genuinely worried? Play along? Take it seriously? Refuse to answer for fear that this is part of a jailbreak for a hacking attempt?
-
Mathematical Singularities and AI Model Validation Failures
By
–
If you eg. use a mathematical model that divides by zero, you've got a singularity. But since mathematics does not allow you to divide by zero, you know that your model is wrong.
-
Stanford scholars publish open-source AI risks benefits analysis
By
–
Open vs. closed AI? Scholars from various organizations including @StanfordHAI and @StanfordCRFM published a paper aimed at creating a more precise understanding of the risks and benefits of open-source AI. (via @axios)https://t.co/j7Nkw5W8xY pic.twitter.com/nodDoWMrVD
— Stanford HAI (@StanfordHAI) 15 mars 2024Open vs. closed AI? Scholars from various organizations including @StanfordHAI and @StanfordCRFM published a paper aimed at creating a more precise understanding of the risks and benefits of open-source AI. (via @axios
) https://
bit.ly/3wTsKDU -
AI Systems Not Yet Ready for Autonomous Error-Free Work
By
–
To be clear, AI systems aren't quite there yet to do this work autonomously and error-free without help. Even afterward, there is a way to go before you would want to trust a major project to AI, but it is a fascinating start nonetheless.
-
AI Watermarking: Minimum Security Standard for Generated Content
By
–
I'm glad to hear you've pulled the project. And true, that kind of watermark can be removed, but that doesn't mean it's not the bare minimum. (And yes, Anthropic et al should also be doing watermarking.)
-
Watermarking Implementation and Ethical Accountability in AI Systems
By
–
Tell me about your watermarks. How are they implemented? Are they both machine & human readable? Why aren't they described in your thread, especially where you are absolving yourself from handling ethical considerations?
-
AI problems evolving from basic issues to nuanced challenges
By
–
First, things became problematic, now the problems are getting nuanced