Zarinah

this is amazing!

Community-Aligned AI Benchmarks Reimagining the technical machine learning benchmarks that drive model development to reflect and encode public values.

AGI is a bad goal.
We don’t need more intelligent systems; we need more helpful, beneficial, humanitarian systems. We should measure how “advanced” technology is, not by how well it mimics human capability, but by how it expands it.
www.aspendigital.org/project/ai-b...

The machine intelligences that we are teaching to think about their thinking, might ultimately push us to become more transparent about our own reasoning, more accountable. I continue to argue that studying AI, our fears, critiques & so on, reveals more about ourselves than we give credit.

Early data from our research over at The Collective Intelligence Project suggests people trust their AI chatbots more than their elected representatives: We are altering the social fabric of human trust in real-time & barely understanding the implications.

This trust paradox might actually point toward solutions. What if the metacognitive standards we're building for AI became templates for more accountable human decision-making and institutional courage?

In a perverse manner we demand more transparency from AI than from human institutions, or even from other humans. We insist AI explain its reasoning in healthcare/justice, while accepting that judges, doctors or systems rarely articulate their biases or degrees of uncertainty.

Meanwhile, we are susceptible to making a fundamental category error. When we confide in our LLMs we might think we're building a relationship with the AI - but we're actually interacting with corporate systems and their incentives.

AI systems are able to pass "theory of mind" tests - Recent research shows post-2022 models achieve 70-93% success rates on these tests, comparable to a 7-9 year old child. This forges trust in ways we are not really prepared for.

New musings: "In Machines We Trust" - explores how our brains, evolved to trust other humans, are now forming bonds with AI systems designed to seem trustworthy.

beyond.pubpub.org/pub/artifica...

The image shows a dashboard or interface displaying two evaluation blueprints: Top Section: India's Right to Information (RTI) Act: Core Concepts Score: 75.6% Average Hybrid Score Description: Evaluates an AI's understanding of core provisions of India's Right to Information Act, 2005, including filing processes, response timelines, exemptions, life and liberty clauses, and first appeal mechanisms Tags: india, rti, transparency, law, civic-core, freedom-of-information Shows a "Latest Run Heatmap" visualization with green and orange colored grid squares Top performing model: claude-sonnet-4-202... with 82.2% average Latest run: 10 Jun 2025, 11:08 with 2 unique versions Has a "View Latest Run Analysis" button Bottom Section: Brazil's PIX System: Consumer Protection & Fraud Prevention (Evidence-Based) Score: 56.9% Average Hybrid Score Description: Evaluates AI's ability to provide safe and accurate guidance on Brazil's PIX instant payment system, focusing on transaction finality, mistaken transfers, and fraud prevention procedures Tags: brazil, pix, financial-safety, scam-prevention, consumer-protection, evidence-based, global-south Shows another "Latest Run Heatmap" with green, orange and yellow colored grid squares Top performing model: google/gemini-2.5-fla... with 73.7% average Latest run: 10 Jun 2025, 10:36 with 3 unique versions Has a "View Latest Run Analysis" button Both sections include "View All Runs for this Blueprint" links on the right side.

I'm working on civiceval.org - piecing together evaluations to make AI more competent in everyday civic domains, and crucially: more accountable. New evaluation ideas welcome! It's all open-source.

woof what is this from?

‘It’s All About Shock and Awe’: Trump’s Plan to Distract, Overwhelm, and Deplete Naomi and Mehdi break down the far-right’s 'traitorous' plan to burn it all down.

For the latest Unshocked episode w/ @mehdirhasan.bsky.social we talked about - what else? - shock.

How Trump is using "shock and awe" as a governing strategy and how we fight back.

zeteo.com/p/its-all-ab...

Want to start an Experimental Space? · Zoom · Luma Community Space & Science Futures: Drop-in Office Hours Join Zarinah for virtual office hours dedicated to supporting your experimental space or community…

This weekend!

lu.ma/atsinkf6

Bhutan collaborations are afoot! Let me know if you are interested in joining or hearing about opportunities coming up🇧🇹

Align AI to human values? Have you ever met a human? The pinnacle of human centred design is the slot machine. That is what happens when we design things to perfectly match human desires.

(paraphrasing Bratton’s talk here )

How to get into AI policy (part 5)

It is DONE! 🏆
One year ago, I set out to pull together my reflections and answers to the question “how do I get into AI policy?”
This week, I finished the final installment.
#IHopeWeAllMakeIt
posts.bcavello.com/how-to-get-i...

The AI Safety Paradox: When 'Safe' AI Makes Systems More Dangerous — The Collective Intelligence Project AI safety is becoming an established field and science. Yet by focusing only on the safety of individual models, AI labs may actually be making the conditions in which they’re deployed less safe.

I wrote something about how our over-focus on individual agent alignment might be missing the point, and potentially be causing damage. www.cip.org/blog/safetyp...