Best-of-N isn't limited to text. We jailbroke vision language models by repeatedly generating images with different backgrounds and overlaid text in different fonts. For audio, we adjusted pitch, speed, and background noise. Some examples are here: https://
jplhughes.github.io/bon-jailbreaki
ng/#examples
…
SAFETY
-
Best-of-N Jailbreaking Techniques for Vision Language Models
By
–
-
Anthropic’s Responsibility Claims: Critique of Big Tech AI Ethics
By
–
I wrote an "open letter" that I decided not to make open back then because how they were being described was infuriating. And Anthropic is doing exactly the same thing now saying that they are the "responsible" ones saving us from OpenAI's bad practices.
-
IndiaAI Calls Safe Trusted AI Projects Expression Interest
By
–
IndiaAI invites Expression of Interest (EoI) from innovators and organizations to lead transformative Safe & Trusted AI Projects that champion accountability, promote fairness, and mitigate harm from AI. Be a part of building a responsible AI future for India!
-
Meta FAIR Releases Nine Open Source AI Research Artifacts
By
–
Wrapping up the year and coinciding with #NeurIPS2024, today at Meta FAIR we’re releasing a collection of nine new open source AI research artifacts across our work in developing agents, robustness & safety and new architectures. More in the video from @jpineau1
. All of this -
Model Self-Improvement: Emergent Behavior in AI Systems
By
–
Check out this example which showcases one of the most exciting research directions: self-improvement.
— Oriol Vinyals (@OriolVinyalsML) 12 décembre 2024
In it, you see this behavior emerging (!) when the model realizes (with a "Oops!") that it did a mistake, and fixes it to create the cute image.
Wild times. https://t.co/bSwfUa7QCD pic.twitter.com/TasWGsgn5sCheck out this example which showcases one of the most exciting research directions: self-improvement. In it, you see this behavior emerging (!) when the model realizes (with a "Oops!") that it did a mistake, and fixes it to create the cute image. Wild times.
-
Teaching Robots Safe Task Execution Using Vision and LLMs
By
–
Can we teach a robot its limits to do chores safely & correctly? 🧵
— MIT CSAIL (@MIT_CSAIL) 12 décembre 2024
To help robots execute open-ended, multi-step tasks, MIT CSAIL researchers used vision models to see what’s near the machine & model its constraints. An LLM sketches up a plan that’s checked in a simulator to… pic.twitter.com/u5HyfQwUicCan we teach a robot its limits to do chores safely & correctly? To help robots execute open-ended, multi-step tasks, MIT CSAIL researchers used vision models to see what’s near the machine & model its constraints. An LLM sketches up a plan that’s checked in a simulator to
-
India AI-Ready: Ethical Frameworks and Deepfake Detection Tools
By
–
Come, join us to make India AI-ready by developing tools, frameworks or solutions across the following themes:
– Watermarking & Labelling
– Ethical AI Frameworks
– AI Risk Assessment & Management
– Stress Testing Tools
– Deepfake Detection Tools -
IndiaAI Launches Safe Trusted AI Projects Initiative
By
–
IndiaAI invites Expression of Interest (EoI) from innovators and organizations to lead transformative Safe & Trusted AI Projects that champion accountability, promote fairness, and mitigate harm from AI.
#AIForGood #ResponsibleAI #SafeAndTrustedAI #IndiaAI #AIFuture #TechForIndia -
AI-powered border surveillance towers and autonomous drones
By
–
D’abord, des tours de surveillance équipées d’IA le long de la frontière américano-mexicaine. Ces tours pouvaient distinguer les humains, les animaux et les véhicules.
Ensuite, leurs systèmes autonomes : Le drone Ghost pouvait voler plus de 15 heures, et l’Anvil traquait et -
Efforts to avoid doom ultimately unsuccessful
By
–
We spent a lot of effort trying to make it not-doom before being pulled back in that direction.