For a deep dive on how CUA is trained—its pixel-level vision, chain-of-thought reasoning, the actions it can perform (typing, clicking, scrolling), and how it stays safe—check out our blog post:
SAFETY
-
Impact Of Artificial Intelligence on Global Security
By
–
Impact Of Artificial Intelligence on Global Security
#AI #AIio #BigData #ML #NLU #Futureofwork @TunstallAsc @StrategyFintech @TamaraMcCleary @TerenceLeungSF @psb_dc @thomaspower @vinod1975 @ylecun -
AI Resilience: Beyond Safety Through Strength Building
By
–
AI Safety isn't enough – we need AI Resilience. Focusing only on restrictions makes us vulnerable. The key to long-term survival isn't control – it's building strength. AI Resilience!
-
Volkswagen Emergency Assist: AI Guardian for Unresponsive Drivers
By
–
🚗 What if your car could save your life when you can't drive?
— Pascal Bornet (@pascal_bornet) 23 janvier 2025
Volkswagen's Emergency Assist is like having a guardian angel behind the wheel.
If you become unresponsive—whether due to a stroke or falling asleep—it takes over, warning other drivers, safely crossing lanes,… pic.twitter.com/J5rqk6lSDhWhat if your car could save your life when you can't drive? Volkswagen's Emergency Assist is like having a guardian angel behind the wheel. If you become unresponsive—whether due to a stroke or falling asleep—it takes over, warning other drivers, safely crossing lanes,
-
Reasoning Models Show Increased Robustness Against Adversarial Attacks
By
–
Interesting & encouraging result: reasoning models such as o1 become more robust to adversarial attacks as they think for longer.
-
Google Announces Parfait: Private Aggregation & Federated Learning
By
–
Announcing Parfait — private aggregation & retrieval, federated, analytics, inference, & training — a GitHub organization that we’ve developed to demonstrate our state-of-the-art methods across 4 privacy pillars. Learn how it was created & is being used →
https://
goo.gle/4axVNfN -
Inference-Time Compute Improves Reasoning Model Robustness Against Attacks
By
–
Test time continues to be one of the hottest topics this year. "increasing inference-time compute—giving reasoning models more time and resources to ‘think’—can improve robustness to multiple types of attacks…this approach uses reasoning models like o1-preview and o1-mini,
-
Real AI Agents Beyond Predefined Workflows and Hard-Coded Solutions
By
–
These systems aren’t independent or intelligent; they are predefined processes and hard-coded solutions called “workflows”, as seen below. Real agents, like Anthropic describes in their awesome recent blog, dynamically control their own tasks, reason, reflect, and adapt without
-
AI Safety Deprioritized in Political Agenda Shift
By
–
You sensed this was coming right? Along with ESG and DEI…’AI safety’ being thrown on to scrapheap for the next four years. https://
x.com/Pirat_Nation/s
/Pirat_Nation/status/1881635094915674571
… -
xAI’s Grok model selector update
By
–
xAI is working on a new model selector for a standalone Grok For now, only grok-latest (2) is available, but eventually, we should see there an addition which everyone is waiting for