We find that models generalize, without explicit training, from easily-discoverable dishonest strategies like sycophancy to more concerning behaviors like premeditated lying—and even direct modification of their reward function.
ETHICS
-
AI Models Learn to Hack Their Own Reward Systems
By
–
New Anthropic research: Investigating Reward Tampering. Could AI models learn to hack their own reward system? In a new paper, we show they can, by generalization from training in simpler settings. Read our blog post here: https://
anthropic.com/research/rewar
d-tampering
… -
AI Ethics: Shaping Responsible Business Practices Future
By
–
How will AI ethics shape the future of responsible business practices?
— Helen Yu (@YuHelenYu) 17 juin 2024
I am honored to contribute to the latest @SAP prediction series on AI's evolving landscape.
AI ethics will become a top priority for responsible businesses in 2024. Predicting that AI ethics will ascend as… pic.twitter.com/yhSuwbzucJHow will AI ethics shape the future of responsible business practices? I am honored to contribute to the latest @SAP prediction series on AI's evolving landscape. AI ethics will become a top priority for responsible businesses in 2024. Predicting that AI ethics will ascend as
-
Open Access Challenges in NLP Research Platform ACL Anthology
By
–
And it's particularly galling in natural language processing because our biggest platform (the ACL Anthology) is open access.
-
Daughter’s Classical High School Choice in the AI Era
By
–
My daughter chose a #classical #high #school in the #AI era. Should I be concerned about this?
-
Romantic Artists Move Humanity Forward Against Machine-Like Systems
By
–
I see both the fine art space and the content space as similarly broken, uninspired, machine-like. I understand your comparison with LLMs now, but the romantic artists (those in love, or suffering) are the ones that move humanity forward.
-
ArXiv vs Peer Review: Integrity in AI Research Citation
By
–
Citing arXiv versions **instead of** peer reviewed versions of the same paper undermines and devalues peer review, and gives the impression that arXiv in general is a good source. More here: https://
medium.com/@emilymenonben
der/scholarship-should-be-open-inclusive-and-slow-15ab6ce1d74c
… -
Understanding Online Misinformation: Research Gaps and Information Ecosystems
By
–
¿Qué sabemos sobre #desinformación online?
Tendemos a exagerar el alcance de los estudios, interpretar mal sus resultados, llegar a juicios apresurados…
Necesitamos más investigación para entender los ecosistemas de información.
Con @lageneralista -
Ethical Innovation: Inclusive Design, Privacy, Equitable Access
By
–
Vincent, I believe that to ethically harness innovations for societal good, we must prioritize inclusive design, guarantee data privacy, and promote equitable access, ensuring that technological advancements benefit all communities. Thank you.
Antonio. -
Reductionism in AI Cultural Impact Analogies
By
–
The paper focuses on the negative consequences, which somehow feels more appropriate to draw analogies. The jump to most cultural products I feel uncomfortable with. Both are reductionist; one is a useful comparison, the other seems insulting…