Related posts

AI researcher at OpenAI; built the poker systems Libratus and Pluribus and the Diplomacy agent Cicero, and works on reasoning models.

• Our readingkorrents.com

  Extra thinking time cannot help with factual recall: a model that does not know a date will not know it after a week either.

  Their words

  if you ask a person when was Abraham Lincoln born and they don't know the date. They could sit there, they could think about it for a week, if they if they don't have access to Wikipedia or something, they're not going to be able to do better answering that question if they thought about it for a week compared to 5 seconds. Same with the model.

  youtube.com

Engineer and teacher; wrote Kubernetes the Hard Way, was a distinguished engineer at Google Cloud, and retired from full-time work in 2023.

• Our readingkorrents.com

  Being an all-star at the activity is not impact: the measure is whether the team's queue is empty, not how many calls you personally took.

  Their words

  And it's the first time I learned the difference between activities and impact. activity is you being an all-star answering a bunch of calls, but the ticket queue for the team is still high.

  youtube.com

Builds developer tools — SST, and now OpenCode at Anomaly. Posts constantly on x.com and writes almost never; the blog stopped in 2021.

• Our readingkorrents.com

  Answering every problem, competitor and user request with one more shipped feature does not add up to a good product; it adds up to a horrible one.

  Their words

  If you add that up, you think, "Oh, we shipped a thousand features. Now that adds it to a good product." It actually adds it to a horrible product.

  Building OpenCode with Dax Raadyoutube.com

Professor at Berkeley working on robot learning; co-founded Physical Intelligence, and argues that robots will learn from data rather than be programmed.

• Our readingkorrents.com

  The goal was never good simulation; it was answering counterfactuals, and a value function does that job as well as a simulator does.

  Their words

  It tells us that the key is not necessarily to do really good simulation. The key is to figure out how to answer counterfactuals.

  youtube.com

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

  People engage more with an argument when they generate its key steps themselves than when the same argument is imposed on them.

  Their words

  It's a lot easier for someone to engage with an argument if they generated the key steps themselves by answering my questions - if imposed by me, it sparks contrarianism and defensiveness

  neelnanda.io

Astrobiologist and theoretical physicist at Arizona State University, working on assembly theory and the physics of life.

• Our readingkorrents.com

  A theory of reality that cannot name the question it is answering is not worth building.

  Their words

  It’s totally fine to have a turtle. Everyone has a turtle. You can’t build a theory without a turtle. It depends on the problem you want to describe. Actually, the reason I can’t get behind Stephen’s ontology is I don’t know what question he’s trying to answer. Without a question to answer, I don’t understand why you’re building a theory of reality.

  youtube.com

Historian of Ukraine at Harvard; wrote The Gates of Europe, Chernobyl and The Russo-Ukrainian War.

• Our readingkorrents.com

  Propaganda is rarely an outright lie — usually it is one real fact taken out of context and blown out of proportion, and answering it on its own terms is already losing to it.

  Their words

  That’s the only way how you deal with propaganda, because propaganda is not necessarily something that is an outright lie. It can be just one factor that’s taken out of the context and is blown out of proportion. And that is good enough.

  youtube.com

Professor at Carnegie Mellon working on self-driving-car safety; writes Safe Autonomy on how autonomous vehicles are and are not shown to be safe.

• Our readingkorrents.com

  The self-driving industry's strategy of answering criticism with 'we are saving lives' has failed, and it will have to change both how it deploys and how it talks.

  self-driving cars

  Their words

  At a higher level, the real event was a catastrophic failure of the industry's strategy of relentlessly shouting as loud as they can "Hey, get off our case, we're busy saving lives here!"

  My Automated Vehicle Safety Prediction for 2024safeautonomy.blogspot.com

Engineer and writer on machine-learning systems; author of Designing Machine Learning Systems and AI Engineering. "I work to bring AI into production. I write about AI system design."

• Our readingkorrents.com

  Text is a more powerful output modality for a model than images, because text supports many downstream tasks

  Their words

  Text is a much more powerful mode for model outputs. A model that can generate images can only be used for image generation, whereas a model that can generate text can be used for many tasks: summarization, translation, reasoning, question answering, etc.

  Multimodality and Large Multimodal Models (LMMs)huyenchip.com

Programmer; founded comma.ai and tinygrad, and was the first to unlock the iPhone.

• Our readingkorrents.com

  Publishing your source code is not open source; open source is a culture of developing in public and answering your issues.

  open source

  Their words

  Just dumping the code on GitHub is not open source. Open source is a culture. Open source means that your issues are not all one year old, stale issues. Open source means developing in public.

  youtube.com