Agents have a WYSIWYG problem. For text, image, and video generation — the generated artifact is something you can look at and at a glance determine quality. For agents, the series of actions they take and the intermediate results is what they produce – often opaque.
SAFETY
-
UK Government Misunderstanding of AI Safety Investment and Regulatory Funding
By
–
Yes, likely both. In the UK case it seems the government (or their advisors) just didn't/don't understand. With 10x more for AISI than the 10 million for the 19 regulators. But it seems odd that same is true throughout Europe … effective lobbying?
-
Boeing safety concerns amid automation integration
By
–
It's not the coming I'm worried about, it's the Boeing.
-
Legal Battles Shape Future Economy of Generative Music AI
By
–
Continúan sumándose batallas legales, en este caso al ecosistema de las IAs generativas musicales. Muchas de las sentencias que salgan de aquí serán las que acaben dando forma a la economía post-IA.
-
Warning Labels for LLMs: Protecting Children from Anthropomorphic AI
By
–
I support @Surgeon_General Vivek Murthy’s reco that SMedia should have warning labels for kids. I also believe we should have these for any LLMs using anthropomorphic language (as most all do). @GaryMarcus Thoughts? Inferring AI is sentient w/out disclosure harms kids & agency.
-
Global poverty, AI, healthcare, and US economic indicators analysis
By
–
Global poverty, global income and wealth, global health, AI and computer power, vaccines and healthcare more broadly, real wages in the United States, crime in the United States, wages of working class people in the United States, inflation in the United States, particulate
-
Goldfish Loss: Mitigating Memorization in Large Language Models
By
–
6/ Mitigating Memorization in LLMs – presents a modification of the next-token prediction objective called goldfish loss to help mitigate the verbatim generation of memorized training data.
-
Future warfare with million-drone squadrons will dwarf 20th century wars
By
–
The wars of the 20th century will look quaint in the days of million-drone squadrons.
-
Tinder’s Production Safety and Trust Systems Using Lorax Technology
By
–
Here’s an upcoming talk from Tinder about their use of Lorax for their production safety and trust systems: https://
ai.engineer/worldsfair/202
4/schedule/ai-frontiers-in-trust-and-safety-combatting-multifaceted-harm-on-tinder-at-scale
….