Far from the utopia promised by the founders of AI labs, global politics is barreling toward an extremely dangerous phase. Make no mistake—I remain profoundly optimistic about AI's potential to solve humanity's greatest challenges, from disease to climate change. The technology
SAFETY
-
AGI Safety Through Iterative Deployment and Continuous Learning
By
–
We’re sharing how we think about safety and alignment. We see AGI development as a continuous path rather than a sudden leap. Instead of preparing for a single pivotal moment, we're iteratively deploying and learning from today's models to make future AI safer and more
-
AI Tools for Legal Access: Addressing Bias and Hallucinations
By
–
75% of 20M annual civil cases involve at least one party without legal representation. AI tools could help, but must be developed responsibly to avoid risks like bias & hallucinations. @StanfordLaw
's David Engstrom explains the potential of courthouse AI. -
AGI Hype Overlooks Basic Reinforcement Learning Gaps
By
–
Amidst all the OMG AGI IS HERE hype, it is amusing to reflect on how some of the most basic continuous online reinforcement learning problems that trivial organisms solve remain unworkable for today’s systems.
— John Carmack (@ID_AA_Carmack) 5 mars 2025
Happy to be working with Rich! https://t.co/7VKpj6wmm2Amidst all the OMG AGI IS HERE hype, it is amusing to reflect on how some of the most basic continuous online reinforcement learning problems that trivial organisms solve remain unworkable for today’s systems. Happy to be working with Rich!
-
Hoping for AGI to save humanity from self-destruction
By
–
I am almost at the point where I think: let AGI or ASI come quickly to save humanity from itself.
-
AI Medical Evaluations Need Real Patient Data Framework
By
–
Only 5% of AI medical evaluations use real patient data, study finds. A new framework adapted from @StanfordCRFM
's HELM aims to bridge the gap between theoretical and practical AI performance in healthcare. @StanfordMed -
Confidential Federated Analytics Enhances Data Privacy Transparency
By
–
Confidential Federated Analytics is a new technique to improve data privacy by publicizing the analyses that will be performed on device data, enabling organizations to be transparent about the privacy properties of their data processing. Read more at https://
goo.gle/3DgbPih -
Deepfake Face Swapping Model Example Demonstrated
By
–
Here's another example I tried with a deepfake model which is used for face swapping.
-
Deepfakes: Revolutionizing Film While Raising Misinformation Risks
By
–
The implications of deepfakes are both exciting and concerning. From revolutionizing the film industry to creating potential misinformation risks, this technology demands responsible use and critical thinking skills from all of us.
-
Deepfake Technology: AI Face Swapping and Voice Synthesis
By
–
We finished the week exploring deepfake technology – AI that can convincingly place someone's face onto another person's body in videos or generate realistic speech mimicking someone's voice.