Unfortunately, SB 1047 has passed the vote despite many AI experts like @drfeifei @ylecun and myself writing detailed articles on how fundamentally flawed it is. @GavinNewsom Please read our letter signed by @Caltech personnel, alumni, and others and veto this.
SAFETY
-
COAR Method Improves Attribution in Vision Language Models
By
–
Experiments on large-scale vision & language models show COAR yields accurate component attributions, outperforming prior methods. This makes it a valuable tool for understanding & debugging complex ML systems.
-
COAR: Component Attribution for Targeted Model Edits
By
–
COAR works by estimating the contribution of each component to the model’s final output. These component attributions act as a counterfactual estimator, enabling targeted model edits w/o additional training.
-
Improving Landing System Reliability Through Iterative Failure Analysis
By
–
Now we figure out what went wrong to drive the landing failure rate far above 1 in a thousand, then 1 in 10 thousand … 1 in a milion, etc
-
Safety Priority: Systems Check Before Launch Confirmation
By
–
We could have launched and it would have been fine, but safety is paramount, so better to check all systems again
-
Grok silently purging stopped responses causes confusing conversations
By
–
Silently purging stopped responses from Grok’s context is confusing — it leads to conversations like this when you stop it because it’s misunderstanding you and you try to clarify:
-
Legality versus ethics in AI: navigating the fuzzy boundary
By
–
Good point… it should at the very least be legal, and ideally ethical, but the latter is a bit fuzzier
-
Research: Humans Outsmart Automated LLM Defenses
By
–
Humans, noted virtuosi of adversarial yap, remain #1 at trolling LLMs! New research from @scale_AI
's SEAL team shows human red teamers achieve 70%+ success rates against LLM defenses that stump automated attacks, exploiting their susceptibility to multi-turn jailbreaks. -
AI Safety in Regulated Industries with Chainguard Partnership
By
–
We're honored to help our customers in highly regulated industries deliver core business value with AI, and do so safely with the assistance of partners like @chainguard_dev . https://t.co/57YlH0ByDu
— Domino Data Lab (@DominoDataLab) 27 août 2024We're honored to help our customers in highly regulated industries deliver core business value with AI, and do so safely with the assistance of partners like @chainguard_dev .
-
Jefferson Lab Enhances Accelerator Safety System with Model-Based Design
By
–
Building a Safer Accelerator with Model-Based Design Read how Jefferson Lab upgraded its Personnel Safety System for the Continuous Electron Beam Accelerator Facility with Simulink, achieving a major efficiency milestone https://
spr.ly/6014lyKwY