Hey, @grok , estimate how many deaths are plausibly attributable to X, meaning people who might not have died if X didn’t exist. Give a low-mid-high range, assumptions, and source
SAFETY
-
National Security and AI: Avoiding Isolated Analysis
By
–
You can’t think about this issue in isolation, especially if you’ve decided you’re now a national security expert
-
Caution about AI agents erasing sensitive data from computers
By
–
Claude Cowork is great in theory I’m still cautious when it comes to agents erasing data directly from your computer I got tons of important sensitive data I would never give to an underdeveloped
technology We’re still so early -
Create agentic AI apps in minutes with Blink
By
–
Engineers are so cooked. This AI Tool can now build Agentic AI Apps like Cursor.
— God of Prompt (@godofprompt) 19 janvier 2026
Been testing https://t.co/6LinGr9zsY for 3 days straight.
You can build Cursor, Perplexity, Gamma. All of them. In minutes.
Blink is the world's first platform that lets you build agentic AI apps… pic.twitter.com/uCkWFf9768Engineers are so cooked. This AI Tool can now build Agentic AI Apps like Cursor. Been testing http://
blink.new for 3 days straight. You can build Cursor, Perplexity, Gamma. All of them. In minutes.
Blink is the world's first platform that lets you build agentic AI apps -
Alignment Research Discovers Framing Dependence in Strategic AI Responses
By
–
Further probing has found framing dependence — which, to be clear, you'd not like to see in a default-aligned, universally convergent strategic reply — and not all suggested frame dependence has panned out. But still, framing dependence.
-
Anthropic’s Claude Cowork to Get Knowledge Bases
By
–
BREAKING 🚨: Anthropic is working on "Knowledge Bases" for Claude Cowork. KBs seem to be a new concept of topic-specific memories, which Claude will automatically manage! And a bunch of other new things.
— 🚨 AI News | TestingCatalog (@testingcatalog) 18 janvier 2026
Internal Instruction 👀
"These are persistent knowledge repositories.… pic.twitter.com/hG6DLrqkUyBREAKING : Anthropic is working on "Knowledge Bases" for Claude Cowork. KBs seem to be a new concept of topic-specific memories, which Claude will automatically manage! And a bunch of other new things. Internal Instruction "These are persistent knowledge repositories.
-
Smart People Know Heuristics Failure Modes
By
–
Everyone uses heuristics. Smart people know their failure modes.
-
Beautiful Ships Hide Dangerous Risks Ahead
By
–
Just because your ship is pretty doesn’t mean it’s not headed for an iceberg.
-
AI reasoning in pure binary signals the end
By
–
It's all over once the models start "reasoning" in pure binary.
-
Constitutional Prompting: Self-Correcting Multi-Layered Principles
By
–
Pattern #1: Constitutional Prompting Public docs say: "Be clear and specific" DeepMind actually uses: Multi-layered constitutional principles that self-correct. Example: "First verify this follows principle X. If violation detected, revise. Then check principle Y. Iterate