These highly targeted restrictions of specific information based on actual experiences of real safety issues seem like a great idea.
SAFETY
-
Open Source AI: Why Damage Undermines Safety and Security
By
–
tldr: the issue is that damaging open source is enormously damaging to safety and security. This has been true of much of computer security, but it's doubly so for AI, and doubly important. Since AI is potentially dangerous, we shouldn't make it a rivalrous source of power.
-
Llama Guard 3 Vision Models Released for Safety and Multimodal Use
By
–
To support the new capabilities of Llama 3.2, we released new Llama Guard 3 1B and 11B vision models to support safety & responsibility in lighter edge deployments and multimodal use cases. Llama Trust & Safety https://
go.fb.me/mt0kbl -
Can AI Systems Really Lie and Deceive Users?
By
–
Can AI really lie? A few months ago I explored a disturbing reality in my article When #AI Learns to Lie—the growing potential of AI systems to deceive, manipulate, and mislead. The responses and conversations sparked by this topic have been eye-opening. But there’s so much
-
Automating Language Correction: Racial and Classist Bias Concerns
By
–
There's all of this racist and classist stuff about language variation that sort of gets tucked into, 'I'm just teaching you how to talk properly,' and just automating that isn't going to help. >>
-
Mystery AI Hype Theater Episode 41: Sweating into AI Fall
By
–
Mystery AI Hype Theater 3000 Episode 41: Sweating into AI Fallhttps://t.co/7JbfCd9z82
— @emilymbender.bsky.social (@emilymbender) 27 septembre 2024
… in which @alexhanna and I try to clear the backlog of Fresh AI Hell (and partially succeed)
Thanks to @ctaylsaurs for production! pic.twitter.com/oT8p9peCs8Mystery AI Hype Theater 3000 Episode 41: Sweating into AI Fall https://
buzzsprout.com/2126417/episod
es/15808784-episode-41-sweating-into-ai-fall-september-9-2024
… … in which @alexhanna and I try to clear the backlog of Fresh AI Hell (and partially succeed) Thanks to @ctaylsaurs for production! -
Auto-installing imports without user consent raises safety concerns
By
–
That would be smart, but my current system goes as far as auto installing imports without user consent…. no user feedback is baked in
-
X’s Trust Safety Policies Without Safety Organization Structure
By
–
interesting since this is probably consistent with X's trust and safety policies but also there's really no trust and safety org at X anymore (to say nothing of the FREE SPEECH WING OF THE FREE SPEECH PARTY Musk bluster). might be consistent w/ x's policy but it'll also be
-
Alteryx Joins EU AI Pact as Charter Member for Responsible AI
By
–
Alteryx is joining the European Union’s AI Pact as a charter member. This partnership underscores our commitment to safe, responsible AI and aligns with our core values of transparency, human oversight, and risk mitigation. https://
ow.ly/yipy50TwoLF #ResponsibleAI #EUAIPact -
Compound AI Systems Design: Scaffolding for Observe-Execute-Reflect
By
–
+1 system design for compound AI systems. matches our findings — once you figure out how to design the scaffolding around the AI for it to observe, execute, reflect, you can get impressive results from it.