Related posts

Chief scientist at Redwood Research, where he works on technical AI safety and AI control.

• Our readingkorrents.com

  Deep properties of a model are inherited through the previous generation's data, which is why AI systems from different companies end up correlated with one another.

  Their words

  And so there's some like deep underlying properties of the model that are being sort of transferred between model generations because basically you you train your AI on data from the prior generation and keep going.

  youtube.com

Professor of economics at George Mason University and Bartley J. Madden Chair at the Mercatus Center. Co-writes the blog Marginal Revolution and co-authors the textbook Modern Principles of Economics with Tyler Cowen.

• Our readingkorrents.com

  Under binding ceilings, cost-minimizing suppliers drive allocations to corners rather than Harberger or random benchmarks, producing larger losses and discontinuous welfare jumps.

  benchmarks

  Their words

  We show that equilibrium generically occurs at neither the Harberger nor Glaeser-Luttmer benchmark. Cost-minimizing suppliers drive allocations to vertices, not interiors. Corners are not an assumption but an outcome about what cost-minimizing suppliers choose. The correct benchmark is corners, not random, and corners generate qualitatively different welfare properties: losses far larger than either efficient or random distributions, and discontinuous jumps from small parameter perturbations.

  arxiv.org

Runs the mechanistic interpretability team at Google DeepMind; worked on interpretability at Anthropic under Chris Olah before that, and teaches the field to newcomers.

• Our readingkorrents.com

  Complete reverse-engineering of neural networks is a less promising research direction than studying model biology and applying interpretability usefully

  Their words

  I'm now fairly pessimistic about ambitious interpretability (i.e. complete reverse-engineering), and I'm excited about model biology (studying qualitative high-level properties of models) and applied interpretability (rigorously doing useful things with interp).

  MATS Applications Open (Due Aug 29)neelnanda.io

Physicist and science communicator. Writes Backreaction and makes videos explaining new physics and astronomy results, frequently to argue that the headline about them overstates the case.

• Our readingkorrents.com

  Violations of Bell inequalities reveal only properties of models, not of nature itself.

  Their words

  Since Local Causality is a property of a model, observing violations of Bell's inequality cannot tell us anything about nature itself, but merely about the properties of the model that we use to describe nature.

  Quantum Confusions, Cleared Up (or so I hope)arxiv.org

Software engineer; co-founder and former CTO of Tailscale, previously on the Go team at Google.

• Mixed onrcmnd.app

  Debian

  Their words

  Debian is a decent programming environment. It has some of the consistency properties like macOS, but fails unexpectedly.

  usesthis.com

Founder and CEO of Get Lighthouse, a management-coaching software company, and previously a product lead at KISSmetrics. Writes about leadership, management and product at jasonevanish.com.

• Likedaffiliate linkrcmnd.app

  Biomimicry: Innovation Inspired

  Their words

  Who are the best engineers and designers in the world? Nature. This book is an interesting, science-heavy look at some really interesting properties in nature we could use to learn from as we try to design new technology. Benyus takes us across the spectrum from photosynthesis, to agriculture, to spiders. If you want to geek out on a little biology while applying it to technology this is a good read.

  jasonevanish.com

Engineering leader and writer; wrote An Elegant Puzzle, Staff Engineer and The Engineering Executive's Primer, and blogs at Irrational Exuberance.

• Our readingkorrents.com

  Organisational fixes are slow because they have to drain years of accumulated state, and that same slowness is what makes them durable once they take.

  Their words

  This is because systems accumulate months or years of state, and you have to drain that all away. Conversely, the same properties that make them slow to fix make them extremely durable once in effect!

  Staying on the path to high performing teams.lethain.com

Game developer and moral philosopher; made QWOP, GIRP and Getting Over It, and teaches at the NYU Game Center.

• Our readingkorrents.com

  Maze puzzles are unengaging on their own but become interesting when paired with a strong extrinsic motivator.

  Their words

  Mazes are not the most interesting gameplay element if you're relying on intrinsic motivation - they're just not very entertaining - but if there's some sufficiently compelling extrinsic motivator I think their irritating properties become quite interesting.

  Hidden Homework (updated 2021)foddy.net

Designer, illustrator and writer; author of The Shape of Design and of widely read essays on design and the web.

• Our readingkorrents.com

  The legibility of the source is one of the most important properties of the web, because it keeps the door open to the next generation.

  Their words

  Illegibility comes from complexity without clarity. I believe that the legibility of the source is one of the most important properties of the web. It’s the main thing that keeps the door open to independent, unmediated contributions to the network.

  Everything Easy is Hard Againfrankchimero.com

Philosopher and historian of existential risk; wrote Human Extinction, and coined "TESCREAL" with Timnit Gebru for the cluster of ideologies they argue underpins the AI industry.

• Our readingkorrents.com

  Never project human mental properties onto an artificial intelligence; the difference between the two architectures is the reason to worry, not a reason not to.

  AI alignment

  Their words

  One of the most commonly emphasized points in the AI safety literature is that one must never anthropomorphize AI—that is, don’t project human mental properties onto artificially intelligent systems.

  thebulletin.org