🎉 Excited to share that our paper, DLBacktrace: Model-Agnostic Explainability for Deep Learning Models, is accepted at #IJCNN 2025! DLBacktrace helps explain how deep learning models make decisions — model-agnostic, efficient, and robust. #InterpretableAI #XAI #TrustworthyAI
SAFETY
-
Open vs Closed Systems: Predictability and Observer Influence
By
–
My point is that open systems are indistinguishable from closed systems in which the influence of the starting state on the observer cannot be predicted in advance
-
Learning Reasoning Without External Rewards in AI Systems
By
–
Learning to Reason without External Rewards
Paper: https://
arxiv.org/pdf/2505.19590
.pdf
…
Code: https://
github.com/sunblaze-ucb/I
ntuitor
… -
Tesla FSD’s Patience With Mom’s Driving Criticism
By
–
The most impressive thing that Tesla FSD can do that I’ve never seen any human is that it stays calm and doesn’t get upset after my mom criticizes its driving for the whole trip
-
AI Assistants Growing More Personal and Creating Deeper Human Attachment
By
–
Relationships and reliance on AI:
— Rowan Cheung (@rowancheung) 27 mai 2025
Demis Hassabis thinks people may start becoming more attached to AI assistants as they increasingly get more personalized.
As they continue to become more powerful and useful, new technologies will be needed. pic.twitter.com/ZoeceXN4MjRelationships and reliance on AI: Demis Hassabis thinks people may start becoming more attached to AI assistants as they increasingly get more personalized. As they continue to become more powerful and useful, new technologies will be needed.
-
DeepTeam: Open-Source Framework for LLM Security Testing
By
–
Test and detect security issues in your LLM Apps! 100% open-source and locally DeepTeam is an open-source LLM red teaming framework to safety test LLM systems. It works with any LLM system, including RAG pipelines, chatbots, and AI agents.
-
Third-party framework proposed for AI flaw reporting accountability
By
–
AI systems are rapidly reaching hundreds of millions, but the infrastructure to identify and fix flaws lags behind. Experts propose a new framework that allows third-party users to report flaws and monitor AI developers' responses and resolutions.
-
Sergey Brin: AI Models Perform Better When Threatened
By
–
Le cofondateur de Google, Sergey Brin : « C’est étrange… on n’en parle pas beaucoup dans la communauté IA… mais en général, tous les modèles — pas seulement les nôtres — ont de meilleurs résultats quand on les menace. » Sacrée dérive…
-
LLMs compared on explaining Tesla FSD RL architecture and safety
By
–
2. Multi-Step Reasoning
— God of Prompt (@godofprompt) 27 mai 2025
Prompt: Explain Tesla FSD’s RL architecture and find edge-case safety flaws (check full prompt in video)
Result:
Gemini: 2
DeepSeek: 0
→ Gemini delivered better logic and depth. DeepSeek R1 fell short. pic.twitter.com/ZpH4gdMxvS2. Multi-Step Reasoning Prompt: Explain Tesla FSD’s RL architecture and find edge-case safety flaws (check full prompt in video) Result: Gemini: 2
DeepSeek: 0 → Gemini delivered better logic and depth. DeepSeek R1 fell short. -
Google Developing AI Studio for Computer Use Apps
By
–
BREAKING : Google is working on a dev environment to let you build Computer Use apps on AI Studio. CUA optimised model soon?