Swaminathan Gurumurthy
2 followers · 43 following · 1147 views
on the atlas · 86
- Reward Hacking in Reinforcement Learning | Lil'Log12 savers
- Base Power Company - Not Boring by Packy McCormick5 savers
- When To Do What You Love33 savers
- Niklas Lauffer - Publications1 savers
- The Iran Opportunity: What America Needs to Do to Achieve a Breakthrough1 savers
- An agenda to maximise AI’s benefits and minimise harms, by David Patterson1 savers
- Approval Extraction Advertised as Production | Compass Rose3 savers
- NVIDIA and Microsoft Boost AI Innovation for Healthcare Startups | NVIDIA Blog1 savers
- Dario Amodei — Machines of Loving Grace50 savers
- Rediscovering the Pleasures of Pluralism: The Potential of Digitally Mediated Civic Participation — Digitalist Papers2 savers
- How China keeps a hold on the supply chain of critical minerals used for EVs and AI : NPR1 savers
- Paul Graham on X: "When To Do What You Love: https://t.co/kMDf3RYhW9" / X1 savers
- Longshoremen at key US ports threatening to strike over automation and pay | AP News1 savers
- How to Fix Defense Procurement - by Santi Ruiz - Statecraft1 savers
- Governance of AI, with AI, through deliberative democracy1 savers
- Leveraging AI for democratic discourse: Chat interventions can improve online political conversations at scale2 savers
- Bad and good arguments for industrial policy1 savers
- Books that pay off — Harvard Gazette1 savers
- Opinion | The Tragedy at This Year’s Hajj Is Just the Beginning - The New York Times1 savers
- Nadia Asparouhova | How to do the jhanas16 savers
- Incepto House3 savers
- No, trade surpluses aren’t caused by comparative advantage1 savers
- rishi's garden - principles3 savers
- We need network societies, not network states — The Collective Intelligence Project1 savers
- AI 2030 | Encode Justice1 savers
- Team | Ought1 savers
- The Psychological Drivers of the Metacrisis: John Vervaeke Iain McGilchrist Daniel Schmachtenberger - YouTube2 savers
- San Francisco AI simulated party: Is this the future of partying?2 savers
- When to Give Up - Ideas Matter6 savers
- Inside TSMC’s Phoenix, Arizona expansion struggles - Rest of World4 savers
- The End of Industrial Society4 savers
- Tech Strikes Back — The New Atlantis4 savers
- Speed matters: Why working quickly is more important than it seems « the jsomers.net blog69 savers
- How to Do Great Work69 savers
- Productivity - Sam Altman58 savers
- Things You Learn Dating Cate Hall - by Sasha Chapin52 savers
- Taste is a guide for what is worthwhile49 savers
- Andy Matuschak45 savers
- Conviction is the Scarcest Resource: Essays About Gaining Real Conviction45 savers
- Conviction is the Scarcest Resource: Dialogues About Conviction32 savers
- How to hire low experience, high potential people31 savers
- Ink & Switch30 savers
- Requests for Startups27 savers
- Do I Need to Go to University? -- colah's blog26 savers
- How to be More Agentic - by Cate Hall - Useful Fictions26 savers
- The Chemistry Of Game Design26 savers
- A List of Interesting Questions for People to Get to Know Their Friends, Family, Lovers, Coworkers, Nemeses, and Selves Better - Google Docs26 savers
- How to Learn Better in the Digital Age25 savers
- Principles of Effective Research | Michael Nielsen23 savers
- The future belongs to those who prepare like Dwarkesh Patel | Meridian22 savers
- The Suicidal Founder - by Anthony Tan - Reboot20 savers
- By default, capital will matter more than ever after AGI — LessWrong19 savers
- how to coparent baby ideas - by Anson Yu18 savers
- Nadia Asparouhova | Explaining tech's notion of talent scarcity17 savers
- Six (and a half) intuitions for KL divergence - LessWrong15 savers
- the agony of eros: dating - by Ava - bookbear express14 savers
- San Francisco, I love you?13 savers
- The Most Dangerous Writing App12 savers
- Give your friends a chance to abandon you - by Kasra12 savers
- Why Ed-tech Startups Don't Scale11 savers
- How to waste your career, one comfortable year at a time11 savers
- Choose Boring Technology10 savers
- My sci-fi novel recommendations - by Noah Smith10 savers
- The Matching Problem in Dating - Erik Torenberg9 savers
- About me | devonzuegel.com9 savers
- Bottlenecks of Aging — Amaranth Foundation8 savers
- The Art of Ideation, Part 1: Be a BUM — gaganbiyani.com8 savers
- Nadia Asparouhova | Reimagining the PhD8 savers
- Semantic reconstruction of continuous language from non-invasive brain recordings | Nature Neuroscience7 savers
- How to find your zone of genius. The saying goes: “If you do what you… | by Mathilde Collin | Medium7 savers
- Gathering Structures7 savers
- The Kool Aid Factory :: Writing In Public, Inside Your Company7 savers
- Winnie Lim » The power of your writing6 savers
- 15 principles for managing up - Wes Kao's Newsletter6 savers
- Why I quit working at an AI startup to join the Department of Energy6 savers
- Top Posts — Neel Nanda6 savers
- The ultimate guide to willingness-to-pay5 savers
- To Invent the Future, You Must Understand the Past | by Leslie Berlin | Backchannel | Medium5 savers
- Towards Small-Scale Social5 savers
- Diffusion models from scratch4 savers
- Don’t Expect to Fix the Climate From a Cushy Office | Lowercarbon Capital4 savers
- Ideaflow3 savers
- The Real Problem at Yale Is Not Free Speech3 savers
- Making Kin with the Machines · Journal of Design and Science3 savers
- 35 Impactful Questions Managers Should Ask Themselves Regularly3 savers
- Startups Roll Out More Services For Growing Ranks Of Self-Employed And Gig Workers2 savers
highlights · 432
Welfare Diplomacy: Benchmarking Language Model Cooperation
Niklas Lauffer - PublicationsTo get more specific on where I think acceleration is likely to come from, a surprisingly large fraction of the progress in biology has come from a truly tiny number of discoveries, often related to broad measurement tools or techniques12 that allow precise but generalized or programmable intervention in biological systems
Dario Amodei — Machines of Loving GraceI believe that in the AI age, we should be talking about the marginal returns to intelligence7, and trying to figure out what the other factors are that are complementary to intelligence and that become limiting factors when intelligence is very hig
Dario Amodei — Machines of Loving Graceany successful engagement infrastructure must give us rights over these co-generated data and relationships, including not just privacy rights but also ownership and use rights.
Rediscovering the Pleasures of Pluralism: The Potential of Digitally Mediated Civic Participation — Digitalist Paperswe need new environments that enable us to tolerate disagreement and difference, that depersonalize conflict, and that enable us to step away when we need to.
Rediscovering the Pleasures of Pluralism: The Potential of Digitally Mediated Civic Participation — Digitalist Paperspreserving human agency and autonomy so that AI and other technologies are not used to manipulate us; encouraging mutual respect; promoting equality and inclusiveness (issues that Tocqueville and his contemporaries passed over); and augmenting rather than substituting for the actions we take as citizens
Rediscovering the Pleasures of Pluralism: The Potential of Digitally Mediated Civic Participation — Digitalist Paperswho “owns” the space—who has agenda-setting power and chooses the topics and questions for collective consideration
Rediscovering the Pleasures of Pluralism: The Potential of Digitally Mediated Civic Participation — Digitalist Papersdynamic hierarchy Leadership is a behavior, not a title or role. Each of us has the capacity to lead, and I believe our society needs to be led by those who are typically in the back — at least occasionally. I thrive in organizations where my leadership is earned not by clout or politick, but rather the merit of what I have to say as a leader. And I believe every leader (including myself) has the responsibility to step back and make space for someone else. When this works, it creates the most inclusive spaces I’ve ever seen.
rishi's garden - principlesWhatever solution our civilization might find to escape the post-industrial trap, it will require social technologies of production and knowledge very different from anything we’ve seen before. A good place to start would be a new basis for friendship that defeats atomization, and a truthfulness that is compatible with political loyalty.
The End of Industrial SocietyOnly external shocks can break the illusion.
The End of Industrial Societythe adjustment to decline would just be an invisible lowering of expectations.
The End of Industrial SocietyCivilizations first lose subtle social technologies and bases of implicit knowledge at crucial junctures; this process ratchets as elites become less adept at institutional management and design; systems downstream of and less complex than the civilizational core then begin to collapse. Eventually, the entire civilization is fatally vulnerable to outside shocks, such as war with more dynamic societies or even mere natural disasters such as pandemics
The End of Industrial SocietyThis means it cannot even mint the preliminary social capital needed to reindustrialize
The End of Industrial SocietyFactories might persist in post-industrial society, but primarily as local traditions. The self-catalytic process that birthed them is long over. They become strange sociological remnants, as out of place in 21st-century society as monasteries were in 19th-century society. Arguably, this industrial traditionalism is the main mode of production in modern Germany and, to a lesser extent, Japan.
The End of Industrial SocietyEurope and the United States are the worst places to put key elements of the supply chain. They cannot produce compliance with public health measures. On paper, they have the best scientists and administrators, but massive white-collar overproduction means the victory of sharp elbows over sharp minds.
The End of Industrial SocietyThe real upward mobility of dysfunctional agriculturalists to functional industrial workers is replaced by the on-paper upward mobility of functional industrial workers to dysfunctional knowledge economy workers. Despite these measures, the real downward mobility is eventually made apparent.
The End of Industrial SocietyThe marketers were eventually replaced by people who grew up on marketing. Beware, social engineer, of obsoleting your own origins! Yet, they did so in field after field. Not only marketers, but scientists, statesmen, industrialists, politicians, philosophers, and writers shaped ersatz social technology to fill the gaps, but completely failed to guarantee knowledge succession of the generative core of knowledge. The strange spiritual practices, scientific exploration of human psychology, and at times outright ideological cults of the founding cohort gave way to a more shallow type of knowledge…
The End of Industrial Societyyou let go of the illusion of having everything by choosing to hone in on having one thing.
find novelty through commitment - by Isabel - Mind MineCommitment is not the end to exploration and independence—it is a deepening of both, aimed inwards
find novelty through commitment - by Isabel - Mind MineSomething that makes our mind and our intuition hum in unison. Something that feels right. Something that demands commitment
find novelty through commitment - by Isabel - Mind Minebecause when you commit to something, you feel safe. And when you feel safe, you explore
find novelty through commitment - by Isabel - Mind MineCommitment simply refines the spectrum of novelty we experience.
find novelty through commitment - by Isabel - Mind Minethe secret about commitment is: it offers everything that optionality promises, but with more depth and dimension
find novelty through commitment - by Isabel - Mind Minewe scaled the industry horizontally more than vertically
Comments - Plentiful, high-paying jobs in the age of AIComputers are far more powerful than they used to be despite the uneven progress, and yet programmers still manage to write slow code that is far from what's technically possible. That's because the industry keeps spending hardware improvements on making developer's lives easier rather than delivering better results to end users
Comments - Plentiful, high-paying jobs in the age of AIA more comprehensive solution is needed. Everyone seems to agree that the solution is to replace this system of waiting lines and self-done studies with central planning
Three threats to the age of energy abundanceA series of new rule changes allow electricity providers to jump the queue if they’re ready, impose penalties for dropping out of line, penalize providers who don’t complete feasibility studies in a reasonable amount of time, and so on. But even FERC doesn’t think these reforms will be enough to fix things, and points out that a number of places have already adopted rules similar to these reforms — with only modest results.
Three threats to the age of energy abundanceenvironmental review laws and interconnection queues
Three threats to the age of energy abundance"Made in China 2025" is different in multiple respects: 1) It focuses on the entire manufacturing process and not just innovation; 2) It promotes the development of not only advanced industries, but traditional industries and modern services; 3) There is still a focus on state involvement, but market mechanisms are more prominent than in SEI. For example, instead of focusing on top-down, unique domestic technical standards, the attention is on self-declared standards and the international standards system; and 4) There are clear and specific measures for innovation, quality, intelligent manufa…
Made in China 202510 priority sectors: 1) New advanced information technology; 2) Automated machine tools & robotics; 3) Aerospace and aeronautical equipment; 4) Maritime equipment and high-tech shipping; 5) Modern rail transport equipment; 6) New-energy vehicles and equipment; 7) Power equipment; 8) Agricultural equipment; 9) New materials; and 10) Biopharma and advanced medical products.
Made in China 2025We need better HRI approaches to designing new and diverse tasks, robustly collecting data, and refining behaviors: improving and streamlining all the careful design that is required to collect data to train models and behaviors, even before an end user even begins to interact with the final product.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumI am also becoming increasingly convinced that data scaling is largely turning into an HRI problem.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumIf instead of having to spend hours authoring a digital replica of my robot and environment, I could capture my robot’s sensors and simply press “play” to generate possible futures, as we did for instance in Video Language Planning, we could solve both the “sim to real” gap and scene authoring problem in one fell swoop.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumgenerative models are the future of simulation, and that instead of just “better simulators” we should also be chasing how to make generative 3D and video respect not only appearance, but physics and spatial relationships.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Medium“universal” arms are widely deployed, but the economics of making them worthwhile are brutal, and they generally require bespoke tooling and systems integration to make them actually useful.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumThat’s a risky bet, particularly if the cross-embodiment hypothesis turns out to be true and a swarm of cheap robots, as exemplified by experiments like Aloha, UMI, Stretch (and Dobb.E,) Mobile Aloha and Aloha 2, could completely disrupt the field and bring dexterous capabilities to a wide variety of cheaper embodiments.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumone embodiment to rule them all” approach to robot learning, with literally billion dollars of investment going into generalist robots such as Tesla, Figure, 1X, Agility, Unitree, Sanctuary, Apptronik
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumon one hand, we’re seeing that cross-embodied models work amazingly at transferring capabilities across robots, which argues for dialing up the variety of robots, tasks, and taking a diversity-focused approach to the problem.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumwe’re still data bound, and improving the efficiency and diversity of data collection is key.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumMany papers have come up with “pixels to action” models, but for us the transformative approach was RT-1. So much has been happening in recent months in this space, with approaches like Action Diffusion from TRI, Humanoid Transformer from Berkeley, ACT and Octo from Stanford, it has been wonderful to see the explosion in performance and capabilities.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumThe next experiment was to merge perception and actuation, while leaving the planner
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumThe first experiment in this direction attempted to merge perception and planning with PaLM-E
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumneural networks can communicate with each other through high-bandwidth, differentiable representations, why force them to collapse them down to words?
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumclose the loop and fine-tune your Code LM based on interactive prompting and refinement of your control behaviors. This is something we dubbed Language Model Predictive Control
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumOne thing LLMs are really good at is generating code, and at the end of the day, controller software is just that: a piece of code describing a policy. This is where the concept of Code as Policies comes in, the idea that you can prompt an LLM with low-level control APIs and let it describe the actual policy to execute.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumwith AutoRT, we used the LLM to dream up new tasks to perform, which immediately raised the question
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumWhich brings me to the next stage of that journey: let’s have the planner and the perception system both use natural language.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | MediumMaybe the planner doesn’t know in advance what information would be useful for the perception module to provide. Maybe it should be a two-way conversation
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Mediumlanguage has a lot going for it: it’s flexible, it’s interpretable, it can describe things at any level of abstraction you choose
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Medium. I am probably biased, but I’ll single out SayCan as the “aha” moment where the community realized that if we lifted much of planning into “semantic space,” as opposed to trying to do it in geometric space, we could use LLMs to do the job, and inherit all the benefits of their common sense abilities without having to collect any data, craft robot-specific ontologies or build symbolic reasoning engines.
The State of Robot Learning. A partially observed, semi-stochastic… | by Vincent Vanhoucke | Mar, 2024 | Medium