🔥Telegram İfşa

Great talk by Wolf Barfuss for Cooperative AI Foundation - there's a lot of value in bringing together the machine leaning community and the complex systems/multi agent econ fields, and this is currently heavily neglected. youtu.be/gHR6xv3xiqE?si…

AI systems can be powerful but opaque "black boxes" - even to researchers who train them. ⬛ Enter Gemma Scope: a set of open tools made up of sparse autoencoders to help decode the inner workings of Gemma 2 models, and better address safety issues. → dpmd.ai/gemma-scope

The big book by Doyne Farmer is out. I wrote this blurb for it. If you are interested in humanity's impact on the environment, economics, and technology I very much recommend the book.

Foundations: Why Britain Has Stagnated. A new essay by Ben Southwood, Samuel Hughes & me. Why the UK's ban on investment in housing, infrastructure and energy is not just a problem. It is *the* problem. And how fixing it is the defining task of our generation. ukfoundations.co

I’ve been saying for a while that stagnation is a policy choice: the UK can choose to be rich! Sam Bowman Ben Southwood + Samuel Hughes lay out the case in brilliant, horrifying detail. Read it, send it to your MP and ask them what they’re going to do about it

At @AISafetyInst we're researching 'safety cases' - careful arguments that AI systems are safe. Join my new safety cases policy team! You'll write strategy, engage labs and governments, fund cost-effective projects – and get stuck into the research, no ML experience required.

With very rare exceptions, the economics profession is completely discrediting itself by either ignoring or badly misunderstanding AI

When I joined Google DeepMind last year, I came across this incredible group of people working on deliberative alignment, and managed to convince them to join my team in a quest to account for viewpoint and value pluralism in AI. Their Science paper is on AI-assisted deliberation

cope in 2022: "it can't even draw hands" cope in 2024: "it can't even maintain coherence over medium to long time frames when simulating complex environments"

You, a normie: I have no idea when to use o3-mini-high or o1. Me, a connoisseur: the o3-mini-high is really more of a sativa. Very cerebral. Gets those creative juices flowing. o1 is more of a body high. Couch locked, contemplating the universe.

We're looking for strong ML researchers and software engineers. You *don't* need to be an expert on AGI safety; we're happy to train you. Learn more: alignmentforum.org/posts/wqz5CRzq… Research Engineer role: boards.greenhouse.io/deepmind/jobs/… Research Scientist role: boards.greenhouse.io/deepmind/jobs/…

The plan? We’ll start a sparkling wine business out of Champaign, IL

Turns out that Self-Other Overlap (SOO) fine-tuning drastically reduces deceptive behavior in language models—without sacrificing performance. SOO aligns an AI’s internal representations of itself and others. We think this could be crucial for AI alignment...🧵

Excited to share Google DeepMind's AGI safety and security strategy to tackle risks like misuse and misalignment. Rather than high-level principles, this 145-page paper outlines a concrete, defense-in-depth technical approach: proactively evaluating & restricting dangerous

Just released GDM’s 100+ page approach to AGI safety & security! (Don’t worry, there’s a 10 page summary.) AGI will be transformative. It enables massive benefits, but could also pose risks. Responsible development means proactively preparing for severe harms before they arise.

two chainsaws passing in the night

It's going to be a while before I can properly write this one up, and I certainly have plenty of disagreements here, but this is a very, very good paper. It is The Way.

Nathan Fielder embodies "you can just do things" more than almost all people in the Bay Area, sorry

Simulations are the future, & one of the main tools we’ll ultimately use to understand and predict things about the universe. This is why I’m so excited about Genie 3, our latest interactive world simulator - here are some insanely cool things you might have missed about it 🧵:

when an AI researcher tells you they're "working all day" they're not really working. they just take their laptop to the bathroom to check how their runs are doing at 2 am but its not actually work. just a severe gambling addiction