I don’t think we need to be afraid or cause panic. We need to be realists. Weigh the pro’s and con’s and work to mitigate risk, like we generally do as a society at large.
SAFETY
-
Auditing AI Systems: Access Requirements for Effective Oversight
By
–
Auditing AI: How Much Access Is Needed to Audit an AI System? https://
bit.ly/3tN0PnD #AI #MachineLearning #DeepLearning #LLMs #DataScience -
Midjourney urged to address bias in AI systems
By
–
I hope @midjourney does something about this issue to reduce bias in the systems. Thanks @matlg for sharing
-
DALL·E 3 Limitations: Spatial Awareness, Text Generation, Hallucinations
By
–
While undoubtedly impressive, it's crucial to remain aware of DALL·E 3's limitations. The image generation model still struggles with spatial awareness, exact text generation in images, and the captioner is known to hallucinate – or invent – details that are absent in the image.
-
Scale AI Safety Lab Advances Artificial Intelligence Safety
By
–
Read more on forbes: https://
forbes.com/sites/richardn
ieva/2023/11/07/scale-new-ai-safety-lab/
… -
Scale AI launches SEAL safety lab for trustworthy AI evaluation
By
–
We @scale_AI are starting SEAL: a new frontier safety lab led by @summeryue0
—prev Bard RLHF lead at DeepMind. We plan to do frontier research in trustworthy evaluations, red teaming & scalable oversight; and collaborate w/AI labs & governments closely. -
Balancing Speed and Accuracy in Dynamic AI Model Design
By
–
The challenge is to design a model that preserves accuracy amidst dynamic situations, potential threats, and numerous queries, by balancing speedy results with precise forecasts, since users appreciate timely responses, tolerating minor inaccuracies.
-
Google Maps Traffic Data Vulnerable to Manipulation Attacks
By
–
Google Maps uses dynamic, user-sourced data for traffic speed gauging. However, it's prone to manipulation, for example, when a person tricked it into rerouting by walking with a bag full of phones.
-
AI Timestamps Disinformation: Assessing Credibility Threats
By
–
Should we be impressed that it can slap timestamps on complete nonsense disinformation like Tucker Carlson videos ?
-
Mistral 7B in Stuffed Parrot Raises Safety Concerns
By
–
This person has put a Mistral 7B model into a stuffed parrot and displays Chinchilla Equation (2) on his torso. Does anyone have his number? This seems a little unsafe.