Better Alignment with Instruction Back-and-Forth Translation discuss: https://
huggingface.co/papers/2408.04
614
… We propose a new method, instruction back-and-forth translation, to construct high-quality synthetic data grounded in world knowledge for aligning large language models (LLMs). Given
ETHICS
-
Instruction Back-and-Forth Translation for Better LLM Alignment
By
–
-
OpenAI Leader Grapples with Work-Life Balance and FOMO
By
–
A surprisingly hard part of my break is beginning—the FOMO for everything happening at OpenAI right now. Lots of results cooking. I’ve poured my life for the past 9 years into OpenAI, including the entirety of my marriage. Our work is important to me, but so is life. I feel ok
-
Billionaire Tech Founders’ Hidden Influence Through AI Philanthropy
By
–
i don't think it's a secret that a16z is billionaire techie-founder AI-investor $$$! 😀 otoh open philanthropy et al being vehicles for billionaire techie-founder AI-investors is less well understood.
-
AI Research Reveals Media Representation Gaps for Underrepresented Groups
By
–
Growing up, did the media lack characters who looked like you? Our Research team partnered with @GeenaDavisOrg , @IAA__India, and @USC to study representation across media using #AI. Let's create a world where everyone sees themselves reflected on screen. https://t.co/pF2v8p3VC0 pic.twitter.com/FGj5VjKoft
— Google AI (@GoogleAI) 8 août 2024Growing up, did the media lack characters who looked like you? Our Research team partnered with @GeenaDavisOrg , @IAA__India
, and @USC to study representation across media using #AI. Let's create a world where everyone sees themselves reflected on screen. https://
goo.gle/3YBpWqy -
AI Model Voice Cloning Without User Consent Raises Safety Concerns
By
–
Esta es de las más locas. Si no restringes al modelo para que hable con ciertas voces determinadas, es capaz de clonar tu voz. Incluso de forma no solicitada! De repente, empieza a hablar como tú. WTFhttps://t.co/keCMRo6AyO
— Carlos Santana (@DotCSV) 8 août 2024Esta es de las más locas. Si no restringes al modelo para que hable con ciertas voces determinadas, es capaz de clonar tu voz. Incluso de forma no solicitada! De repente, empieza a hablar como tú. WTF
-
GPT-4o Capabilities Limited by Safety Restrictions Analysis
By
–
Leyendo la system card de GPT-4o es curioso ver que la mayoría de cosas que probamos y que el modelo es incapaz de hacer (cantar, detectar interlocutores, imitar ciertos personajes, etc) no las hace no porque no pueda, sino porque la han tenido que limitar mucho por seguridad.
-
Understanding AI System Cards: Documentation for Model Transparency
By
–
Una "model card" o "system card" es una documentación que suelen acompañar a los nuevos modelos o sistemas de IA para dar una descripción detallada de muchos aspectos relacionados con su funcionamiento, ética en el entrenamiento, seguridad, etc. https://
ai.meta.com/blog/system-ca
rds-a-new-resource-for-understanding-how-ai-systems-work/
… -
Impossible Promises: How AI Commitments Defy Mathematical Reality
By
–
How are you planning that we'll be required to make promises that are mathematically impossible to keep?
-
Bill Analysis Requires Understanding Actual Language
By
–
This is shockingly credulous and will age very badly. You should spend more time understanding what the bill actually says, rather than what you want it to say. (Which is hard, because they keep making massive changes to it).