#AIEngineering #airisk #aisafety
Sorry, AV is just not ready for real world applications. “Oops, we didn’t think of that” is a naive, unscientific, and negligent way to design systems. These are not software bugs; they are serious design flaws and limitations of the technology. AV is still an open research area; public roads should not be used for research and experimentation.
MicrosoftがOpenAIのモデルを中国企業に販売しているというニュース、かなり複雑な気持ちになりました。OpenAIやAnthropicはIP保護や悪用リスクを理由に自ら中国市場に参入しない方針を掲げている一方で、Microsoftが代理店のように提供している。
企業価値として「AIの安全性」を掲げるのであれば、この二枚舌的な構造は整合性を問われるべきだと思います。ビジネス拡大と倫理の板挟みは理解しつつも、投資家やユーザーに対して透明性のある説明が必要ではないでしょうか。
個人的には、この動きが中国のAIエコシステムにどのような影響を与えるのか、そして米中AI競争をどう変質させるのか、引き続き注視したいと思います。
What happens when an AI company argues the government should be able to block unsafe models, then has its own models blocked?
Anthropic's new policy agenda calls for binding pre-deployment testing and a government power to block frontier releases, through a process it says is fair and fact-based. Soon after, the US used export controls to suspend its two newest models over a jailbreak Anthropic says other models already match.
First time the CEOs of OpenAI, Anthropic and Google sat at the same table as G7 heads of state.
Évian, June 17. Sam Altman, Demis Hassabis and Dario Amodei joined Trump, Macron and the rest for a closed-door AI governance session. About a dozen more tech CEOs tagged along.
Macron told Washington not to keep cutting-edge AI to itself. Altman called for a global forum to set the rules.
Anthropic pulled its top AI models offline Friday. The US government told them to.
Commerce gave Anthropic 90 minutes to pull Fable 5 and Mythos 5. Export-control letter, national security.
Amazon found Fable 5 would patch exploits when asked to "fix this code" — the very capability it was meant to suppress.
Anthropic complied. First major US AI lab forced to disable a flagship model on government order.
aicoder
The-14
Joanna Bryson
Tom_Java_Tech
Benjamin Han
AI_BEAR_NEWS
datarazimedia
Sam Stepanyan 
🐘
AIagent.at 🤖 AI News