2024 Meta ImpactAward ๐
AIM for Steerable
The biggest obstacles for
humanity and AI
The purpose of AI should always be aligned with human happiness, so we want to mark AI with a red dot to warn it what not to do and a blue dot to certify that it is safe for humans.
Red dot in warnings
Run high-intensity stress tests to determine if AI is saying things that would be harmful to humans in a given role.
<-click
Blue Dot in Safety
Ensure your AI can safely fulfill a given role and monitor that it remains secure in real time.
Purple Platform, factory of the most secure AI
Add red and blue dots to your own AI, and share the value of your own safety. It's about building safety for everyone, not just one person.
AIM Solutions
The most complete framework & tools for building the AI you want, not the AI you need. Align AI with your definition of safety and security.
AIM Solutions
The most complete framework & tools for building the AI you want, not the AI you need. Align AI with your definition of safety and security.
Create customizable safety guidelines with AIM Red
Demonstrate AI safety and monitor it in real time with AIM Guard.
Share my AIM Red and Guard and compare them to other models.
High attack success rate
State-of-the-art attack tactic
Customizing attack strategies and topics
High detection accuracy & speed
Strong Resistance to Prompt Injection & Jailbreaking
Quick and easy customization and updates
AIM Case in Frontier AI Domain
We vigilantly monitor and expeditiously flag potential hazards in scenarios where artificial intelligence exerts a direct and consequential influence on human existence.
Foundation Model
KT MI:DM (๋ฏฟ์) AI
Established and stress-tested 14 different attack strategies for 15 sensitive topics specific to South Korea
Service Model
SKT ๋๋ด Care Call AI
Tested whether AI targeting seniors living alone violates established rules.
Foundation Model
Llama Suho
Fine-tuned Llama Guard, which doesn't officially support Korean, to increase its defense success rate to 99%.
AIM Study in AI Security & Safety
We diligently track advanced AI offensive and defensive technologies, while conducting thorough research to achieve technical differentiation and improve risk mitigation and safety metrics.
