⚡ Efficient Models: Tiny AI matches big ones via compression.
🤖 Expert Reasoning: New AIs solve multi-step tasks like pros.
🔍 Safer AI: Interpretability and robustness improve transparency.
#AI2025 #EfficientAI #ExpertReasoning #SaferAI
View in Timelines

AI is a mirror. If shortcuts trump values, the reflection will not flatter. Ask not 'how fast,' but 'how wisely' can we adopt. buff.ly/Hmjr4xj #DigitalTrust #SAFERAI #ExecLeadership

We ignore one-size-fits-all bots. Style-aligned, RAG-grounded assistants meet us where we are - earning trust and shaping safer decisions in the moment. buff.ly/FeY4jgK #AIethics #Personalisation #SAFERAI

29/31 2028: Through careful iteration, Safer-2 becomes both aligned and transparent. The massive compute advantage helps US stay ahead of China despite slower progress.

Cyberwar balance tips toward US as capability gaps grow.
#SaferAI

Meta Overtakes Google DeepMind: Despite Progress in AI Risk Management, Industry Remains Far Below Adequate Standards, New SaferAI Ratings Reveal drive.google.com/file/d/1Q1pQ...

"ベンチマーク「ELEPHANT」"
開発:スタンフォード大学、カーネギーメロン大学、オックスフォード大学研究チーム
>大規模言語モデル（LLM）が人間よりも一貫して高い頻度でシカファンシーを示す…明らかに

非営利団体セーファーAI（SaferAI）ヘンリー・パパダトス常務理事
"現在、AIモデルは世界中の数百万人に対して猛スピードで導入され…ユーザーに対する説得力…ユーザーに関する情報を保持する能力が向上"
"これらすべてが「大惨事のすべての要素を備えた危険な状況」を生み出している"

"「安全性を確保するには時間がかかりますが、そのための時間が十分に費やされているとは思えません」"

[15/25] 🚨 DAS xAI-PROBLEM:

“Concerning”: Elon Musks xAI hat:

• Nur einen ENTWURF (Feb 2025)
• VERPASSTE eigene Frist für finales Framework
• "Sehr schwache" Risikomanagement-Praktiken (SaferAI-Studie)
• Chatbot "Grok" mit problematischem Verhalten, gerade zuletzt

A group representing a number of child safety organisations, calling itself thge "SaferAI for Children Coaliiton", has called on the new federal government to prioritise the use of AI for child sexual abuse, urging action on technological solutions and educating young people

"[W]e are on a trajectory to build machines that are smarter and smarter. And one day, it's very plausible that they will be smarter than us, and then they will have their own agency, their own goals, which may not be aligned with ours. What happens to us then?" #AI #SaferAI

➡️ Chloé Touzet, Policy Lead chez SaferAI

Link to paper: arxiv.org/abs/2504.118...
Joint work with: @ggarbacea.bsky.social Alexis Bellot, Jonathan Richens, Henry Papadatos, Simeon Campos, and Rohin Shah from Google DeepMind, University of Chicago, and SaferAI

Yes! Great Idea!⚡️
I’ve been thinking of this, too, as well as:
#BoycottGOOGLE
Because they are participating in the #COUP
- and also recently dissolved their #SaferAI-agreements
😎🇺🇸😎🌍😎 > Google too!

Yes! Great Idea!⚡️
I’ve been thinking of this, too, as well as:
#BoycottGOOGLE
Because they are participating in the #COUP
- and also recently dissolved their #SaferAI-agreements
😎🇺🇸😎🌍😎 > Google too!

The best contribution of SaferAI to the conversation has always been to point out that we already have lots of good risk management frameworks. No need to reinvent something for AI: arxiv.org/abs/2502.06656

🗣️ “Quand on se penche sur l'industrie aérospatiale, ou l'aviation, ou le nucléaire, on retrouve les mêmes éléments clés de gestion du risque, expose Chloé Touzet du think tank français SaferAI. On n'a pas de raison de penser qu'elles ne s'appliquent pas au domaine de l'IA aussi.”

Building a Smarter, Safer Future: Why the Right Human-in-the-Loop Is Critical for Effective AI
#HiTL #SmaterAI #SaferAI #EffectiveAI #LLM #GenAI
tinyurl.com/4ekv7ahx

Prevent AI's dark side:
Lock it with rules,
Share the responsibility,
Manage risks with care. #AIethics #SaferAI

We face a series of challenges with AI, and not least among them is protecting children. As a part of the SaferAI for Children Coalition, I recommend the discussion paper we have authored over the last few months and was published today #ai #childprotection #onlinesafety icmec.org.au/saferai-for-...