# imbue.com

> AI-optimized mirror of imbue.com containing 195 pages totalling 434,912 words of clean markdown content, structured data, and semantic HTML. Original source: https://imbue.com. Last updated: 2026-07-18T05:38:26.739Z.

Each page is available as HTML (with JSON-LD structured data) and Markdown (text-only, ideal for LLMs and RAG).

## Homepage

- [Imbue | Substack](/content/ideas/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (107 words)
- [We build honest software - Imbue](/content/site-root.html): We build AI to help people think, create, and build. Tools that are loyal to you, not the company that built them. (230 words)

## Articles & Blog Posts

- [Julian Chibane, MPI-INF: 3D reconstruction using implicit functions](/content/ideas/p/episode-06-julian-chibane-mpi-inf-187/index.html): Julian Chibane is a PhD student at the Real Virtual Humans group at the Max Planck Institute for Informatics in Germany. (9,526 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/228675892/index.html): No disrespect for what is clearly a good-faith, well-intentioned effort here, but I am struck by the disconnect and attendant irony between "a space to think together about the technological future" and the photo of a conversation outdoors in the sunshine around a fire pit. Online "spaces" and webinars and classes, etc., are all wonderful and afford millions opportunities they wouldn't enjoy otherwise, but at the end of the day, what we're losing is the ideal: Humans being *with* other humans in meaningful ways. I want a future where technology is genuinely subordinated to the needs of humans, not the foundation for our "spaces." (711 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/228694110/index.html): I agree with you! We host many in-person events for this very reason, including my Art of Being Human event series. My primary intent for hosting the series is to gather people of different backgrounds, with good food and drink, to discuss the human questions that pertain to all of us; and then to use this Substack to share these ideas with people who are not in SF, but who might appreciate them and feel inclined to share their own thoughts.

We shouldn’t take digital spaces to be an adequate substitute for physical gatherings, but I do think the Internet can be a connective force to converse with people we might not otherwise encounter (like our conversation here!), to be exposed to perspectives outside of our immediate milieu, and to, ideally, ultimately meet in-person. I like to think of digital platforms as a means of correspondence (like the Republic of Letters) rather than a space to reside in, and shudder to think of the virtual “metaverse” that many tech moguls are pushing for.

I believe we share the same hope for technology: to enable us to spend more time with people we love, doing what is best in life — relishing in good conversation, cooking and eating together, being in the sun — not staring at screens. This aspiration is the core of the work we do: creating software and agents that can work reliably on our behalf, that can work without our constant supervision, so the technologies in our lives are tools that we can pick up and put down at our own will, to use toward our own ends, rather than devices we feel beholden to. (598 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/228721312/index.html): I really appreciate this response, thank-you. As I said, the good-faith and good intentions here are obvious and shared.

I was going to acknowledge the in-person events you do there but then realized it sort of goes into the problem deeper, and I didn’t want to get lost in the weeds. Full disclosure: There’s a generational aspect to my response (I’m Gen X) and I’ll confess that depression sometimes “talks over” what I’m trying to say (I’ve acknowledged mental health struggles in my ‘stack, so it ain’t a secret).

I’ve definitely taken advantage of the Internet’s connective force, through classes and webinars, online learning platforms (organic, ground-up) and reading ‘stacks where “my people” are, but those connections pale next to in-person events. I find Zoom conversations, particularly, to be an almost annihilatingly lonely experience.

It’s laudable that you’re doing in-person events there, but that’s 600 miles away. My two favorite stacks regularly do in-person events that I'd love to attend, but one is in Britain and the other is on the East Coast; they might as well be on the moon. Roughly a fifth of Americans live in metropolitan areas, the rest of us are out here in small towns and cities and in the country. A lecture or discussion group or author reading in SF or LA or New York means nothing to me … or rather, it means nothing to the vast majority of people. So we’re left with the online spaces to “reside in” if we want to be connected with kindred spirits and fellow travelers at all.

I don’t know the answer, and I suppose there’s an inevitable irony in the fact that if I did have the answer, all I could do, and would do, is share it online and hope it gets legs. But with a couple hundred subscribers, I doubt that it would. (318 words)
- [ideas/p/software-as-bonsai/index.html](/content/ideas/p/software-as-bonsai/index.html) (1,534 words)
- [Sarah Jane Hong, Latent Space: Neural rendering & research process](/content/ideas/p/episode-02-sarah-jane-hong-latent-b7a/index.html): Sarah Jane Hong is the co-founder of Latent Space, a startup building the first fully AI-rendered 3D engine in order to democratize creativity. (900 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/226350008/index.html): Its great to see Imbue building a space for tech future. We @The LAb are also very invested in this. Our goal is to provide a platform to seamlessly build cutting edge tech skills by working in industry or industry grade projects. All technology problems are human problems and needs collaboration, partnership and learning from each other. Looking forward to participate. (61 words)
- [Yash Sharma, MPI-IS: Generalizability, causality, and disentanglement](/content/ideas/p/episode-14-yash-sharma-mpi-is-on-a39/index.html): Yash Sharma (Google Scholar) (Website) is a Ph.D. (527 words)
- [Jonathan Frankle, MIT: The lottery ticket hypothesis and the science of deep learning](/content/ideas/p/episode-13-jonathan-frankle-mit-on-4d4/index.html): Jonathan Frankle (Google Scholar) (Website) is finishing his PhD at MIT, advised by Michael Carbin. (828 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/226565712/index.html): That’s quite interesting! I do agree that AI will result in a shift in our inner experience — in what ways do you think AI will change our inner lives? (133 words)
- [Yilun Du, MIT: Energy-based models, implicit functions, and modularity](/content/ideas/p/episode-16-yilun-du-mit-on-energy-5bf/index.html): Yilun Du (Google Scholar) is a graduate student at MIT advised by Professors Leslie Kaelbling, Tomas Lozano-Perez, and Josh Tenenbaum. (307 words)
- [Martín Arjovsky, INRIA: Benchmarks for robustness and geometric information theory](/content/ideas/p/episode-15-martin-arjovsky-inria-851/index.html): Martín Arjovsky (Google Scholar) did his Ph.D. (403 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/226621493/index.html): Well, since you asked . . . my entire Substack is trying to answer that extremely important question!

In one post I list 25 "mind changers": ". . . twenty-five transformative breakthroughs that, working together, are already beginning to redesign our deepest experiences of being alive. I've arranged these innovations into four categories, describing mind changers that alter our experiences of 1) self, 2) others, 3) reality, and 4) bodies. Like AI, many of these innovations will transform more than one area of our experience, but I'll separate them here for ease of reading."

https://mindrevolution.substack.com/p/opening-my-cabinet-of-mind-changers (96 words)
- [ideas/p/love-labour-and-language/index.html](/content/ideas/p/love-labour-and-language/index.html) (691 words)
- [Cinjon Resnick, NYU: Activity & scene understanding](/content/ideas/p/episode-03-cinjon-resnick-nyu-on-51e/index.html): And how spending time at circuses and with gymnasts re-invigorated his research (11,623 words)
- [Yujia Huang, Caltech: Neuro-inspired generative models](/content/ideas/p/episode-07-yujia-huang-caltech-on-73b/index.html): Yujia Huang (@YujiaHuangC) is a PhD student at Caltech, working at the intersection of deep learning and neuroscience. (13,483 words)
- [Drew Linsley, Brown: Inductive biases for vision and generalization](/content/ideas/p/episode-09-drew-linsley-brown-on-3b2/index.html): Drew Linsley (Google Scholar) (Website) is a Paul J. (13,755 words)
- [Giancarlo Kerg, Mila: Approaching deep learning from mathematical foundations](/content/ideas/p/episode-08-giancarlo-kerg-mila-on-40f/index.html): Giancarlo Kerg (Google Scholar) is a PhD student at Mila, supervised by Yoshua Bengio and Guillaume Lajoie. (15,582 words)
- [Dylan Hadfield-Menell, UC Berkeley/MIT: The value alignment problem in AI](/content/ideas/p/episode-10-dylan-hadfield-menell-6ed/index.html): Dylan Hadfield-Menell (Google Scholar) (Website) recently finished his PhD at UC Berkeley and is starting as an assistant professor at MIT. (17,602 words)
- [Vincent Sitzmann, MIT: Neural scene representations for computer vision](/content/ideas/p/episode-11-vincent-sitzmann-mit-on-1c3/index.html): Vincent Sitzmann (Google Scholar) (Website) is a postdoc at MIT. (15,004 words)
- [Jacob Steinhardt, UC Berkeley: Machine learning safety, alignment, and measurement](/content/ideas/p/episode-12-jacob-steinhardt-uc-berkeley-dd8/index.html): Jacob Steinhardt (Google Scholar) (Website) is an assistant professor at UC Berkeley. (11,717 words)
- [ideas/p/x-com/imbue_ai/index.html](/content/ideas/p/x-com/imbue_ai/index.html) (628 words)
- [ideas/p/episode-01-kelvin-guu-google-ai-on-510/index.html](/content/ideas/p/episode-01-kelvin-guu-google-ai-on-510/index.html) (993 words)
- [Andrew Lampinen, DeepMind: Symbolic behavior, mental time travel, and insights from psychology](/content/ideas/p/episode-17-andrew-lampinen-deepmind-37b/index.html): Andrew Lampinen (Google Scholar) is a Research Scientist at DeepMind. (730 words)
- [Sugandha Sharma, MIT: Biologically inspired neural architectures](/content/ideas/p/episode-26-sugandha-sharma-mit-on-24e/index.html): Sugandha Sharma is a Ph.D. (599 words)
- [Oleh Rybkin, UPenn: Exploration and planning with world models](/content/ideas/p/episode-18-oleh-rybkin-upenn-on-exploration-14a/index.html): Oleh Rybkin is a Ph.D. (551 words)
- [Jack Parker-Holder, DeepMind: Open-endedness and evolving agents](/content/ideas/p/episode-24-jack-parker-holder-deepmind-e17/index.html): Jack Parker-Holder recently joined DeepMind after his Ph.D. (542 words)
- [Noam Brown, FAIR: Building human-level AIs in poker and Diplomacy](/content/ideas/p/episode-27-noam-brown-fair-on-achieving-3e4/index.html): Noam Brown is a research scientist at FAIR. (561 words)
- [Archit Sharma, Stanford: Unsupervised and autonomous reinforcement learning](/content/ideas/p/episode-22-archit-sharma-stanford-54a/index.html): Archit Sharma is a Ph.D. (481 words)
- [Jim Fan, NVIDIA: Foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant](/content/ideas/p/episode-29-jim-fan-nvidia-on-foundation-7b2/index.html): Jim Fan is a research scientist at NVIDIA and got his PhD at Stanford under Fei-Fei Li. (826 words)
- [Minqi Jiang, UCL: Environment and curriculum design for general RL agents](/content/ideas/p/episode-19-minqi-jiang-ucl-on-environment-5c3/index.html): Minqi Jiang is a Ph.D. (399 words)
- [Hattie Zhou, Mila: Supermasks, iterative learning, and fortuitous forgetting](/content/ideas/p/episode-20-hattie-zhou-mila-on-supermasks-566/index.html): Hattie Zhou is a Ph.D. (475 words)
- [Nicklas Hansen, UCSD: Why algorithms don't drive research progress](/content/ideas/p/episode-25-nicklas-hansen-ucsd-on-cb0/index.html): Nicklas Hansen is a Ph.D. (449 words)
- [Ben Eysenbach, CMU: Designing simpler, more principled RL algorithms](/content/ideas/p/episode-30-ben-eysenbach-cmu-on-designing-f15/index.html): Ben Eysenbach is a Ph.D. (665 words)
- [Chelsea Finn, Stanford: Biggest bottlenecks in robotics and reinforcement learning](/content/ideas/p/episode-21-chelsea-finn-stanford-04a/index.html): Chelsea Finn is an Assistant Professor at Stanford and part of the Google Brain team. (660 words)
- [Jamie Simon, UC Berkeley: Theoretical principles for deep neural networks](/content/ideas/p/episode-32-jamie-simon-uc-berkeley-6be/index.html): Jamie Simon is a fourth-year physics Ph.D. (534 words)
- [Tri Dao, Stanford: FlashAttention and efficient training](/content/ideas/p/episode-33-tri-dao-stanford-on-flashattention-4d3/index.html): Tri Dao is a PhD student at Stanford, co-advised by Stefano Ermon and Chris Re. (632 words)
- [Bill Thompson, UC Berkeley: How cultural evolution shapes knowledge acquisition](/content/ideas/p/episode-31-bill-thompson-uc-berkeley-16a/index.html): Bill Thompson is a cognitive scientist and assistant professor at UC Berkeley. (433 words)
- [Sergey Levine, UC Berkeley: Bottlenecks to generalization in reinforcement learning](/content/ideas/p/episode-28-sergey-levine-uc-berkeley-f5e/index.html): Also: why simulation is doomed to succeed, and how to pick good research problems (21,720 words)
- [Choices and Knives - by glenn mcdonald - Imbue](/content/ideas/p/choices-and-knives/index.html): The collective future we want is a future of kitchens, not vending machines. (1,618 words)
- [Idea Tools for Participatory Intelligence](/content/ideas/p/idea-tools-for-participatory-intelligence/index.html): We need tools that are predicated on our rights, dedicated to amplifying our creative capacity, and judged by how they help us improve our world. (733 words)
- [Comments - 10 theses on attention - by Ashley Zhang - Imbue](/content/ideas/p/10-theses-on-attention/comments/index.html): Reflections from Art of Being Human #2 with philosopher Adam Robbert (207 words)
- [Comments - 10 theses on attention - by Ashley Zhang - Imbue](/content/ideas/p/10-theses-on-attention/comment/230269246/index.html): Great reflections Ashley. I'd push back on thesis #2.

Distraction may be perennial, but the current attention economy works through a qualitatively different mechanism. Variable reinforcement, infinite scroll, and algorithmic feeds do not just compete for conscious focus. They train the nervous system into the same defensive narrowing trauma produces: constriction, shortened time horizons, and collapse of presence.

The monks' complaint about illuminated manuscripts was about beauty pulling the eye. What we face now is engineered dysregulation at population scale.

That distinction changes where the intervention lives. If distraction is perennial, the answer is better discipline or better institutions. If the attention economy is a mass dysregulation system, neither willpower nor legislation reaches the actual site of damage. You can't legislate your way to resonance. You can't detox your way to depth.

I explore this further in The Attention Wound: What the Attention Economy Extracts and What the Body Cannot Surrender

https://yauguru.substack.com/p/the-attention-wound?r=217mr3 (151 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comments/index.html): A space to think together about the technological future we want and how to create it (361 words)
- [Comments - 10 theses on attention - by Ashley Zhang - Imbue](/content/ideas/p/10-theses-on-attention/comment/228019695/index.html): Absolutely love your writing. Point 7 gives me hope and reassurance in building the Quiet Brain Club. (30 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/226542116/index.html): Excited to be a part of this community. My own view is that far too much attention is being given to how AI and other new technologies are changing our external worlds and far too little attention is being give to how they changing our inner worlds--how we experience ourselves and reality itself.

Here's the short version of my view: Human consciousness has passed through four revolutions. The first, more than 100,000 years ago, produced self-awareness. The second produced meta-awareness: the recognition that we could be self-aware. The third, only 5,000 years ago, produced mathematics and writing, making it possible for the first time to transmit knowledge of awareness across generations.

AI is the fourth: not a tool revolution but a mind revolution, one already changing how we experience ourselves and the world.

The real stakes are not our jobs but our inner lives.

But I'm open to being convinced I'm wrong :) (278 words)
- [Comments - Introducing Imbue's Substack - by Ashley Zhang](/content/ideas/p/introducing-imbues-substack/comment/226374716/index.html): I'm pleased to see the introduction of this space for us to share our ideas more publicly with the world! Here's hoping that we can steer the evolution of this technology in a direction that is principled and maximises human wellbeing! (54 words)
- [Comments - From lawless spaces to true liberty: rethinking AI's role in society](/content/ideas/p/matt-boulos/comment/145245014/index.html): JELLY <3 (14 words)
- [Comments - There will be a scientific theory of deep learning](/content/ideas/p/learning-mechanics/comment/249002337/index.html): Very enlightening video! I think the book Understanding Deep Learning kind of goes in that direction (https://udlbook.github.io/udlbook/) (26 words)
- [Imbue | Substack](/content/ideas/notes/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (268 words)
- [Seth Lazar, Australian National University: The political philosophy of AI](/content/ideas/p/episode-34-seth-lazar-australian-185/index.html): Seth Lazar is a professor of philosophy at the Australian National University, where he leads the Machine Intelligence and Normative Theory (MINT) Lab. (1,029 words)
- [Percy Liang, Stanford: How foundation models work](/content/ideas/p/episode-35-percy-liang-stanford-on-68d/index.html): Percy Liang is an associate professor of computer science and statistics at Stanford. (1,394 words)
- [Rylan Schaeffer, Stanford: Investigating emergent abilities of LLMs](/content/ideas/p/episode-37-rylan-schaeffer-stanford-de4/index.html): Rylan Schaeffer is a PhD student at Stanford studying the engineering, science, and mathematics of intelligence. (892 words)
- [Ari Morcos, DatologyAI: Leveraging data to democratize model training](/content/ideas/p/episode-36-ari-morcos-datologyai-e92/index.html): Ari Morcos is the CEO of DatologyAI, which makes training deep learning models more performant and efficient by intervening on training data. (1,182 words)
- [Empowering humans in the age of AI - by Kanjun Qiu - Imbue](/content/ideas/p/empowering-humans-in-the-age-of-ai/index.html): We founded Imbue in 2021 to build an AGI future where humans remain at the helm, shaping powerful AI systems rather than being subordinated to them. (1,159 words)
- [Celeste Kidd, UC Berkeley: Attention, curiosity, and how we form beliefs](/content/ideas/p/episode-23-celeste-kidd-uc-berkeley-9a0/index.html): Celeste Kidd is a professor of psychology at UC Berkeley. (548 words)
- [Archive - Imbue](/content/ideas/archive/index.html): Full archive of all the posts from Imbue. (530 words)
- [Imbue | Substack](/content/ideas/t/podcast/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (250 words)
- [AAI - by glenn mcdonald - Imbue](/content/ideas/p/aai/index.html): The missing thing in Artificial Intelligence is not generality, it's adaptation. (1,650 words)
- [A more radical Imbue - by Josh Albrecht - Imbue](/content/ideas/p/a-more-radical-imbue/index.html): Our experiment in working with radical agency, control, and choice (1,511 words)
- [Joel Lehman, OpenAI: Evolution, open-endedness, and reinforcement learning](/content/ideas/p/episode-04-joel-lehman-openai-on-8e0/index.html): Joel Lehman was previously a founding member at Uber AI Labs and an assistant professor at the IT University of Copenhagen. (16,990 words)
- [10 theses on attention - by Ashley Zhang - Imbue](/content/ideas/p/10-theses-on-attention/index.html): Reflections from Art of Being Human #2 with philosopher Adam Robbert (748 words)
- [Introducing Imbue's Substack - by Ashley Zhang - Imbue](/content/ideas/p/introducing-imbues-substack/index.html): A space to think together about the technological future we want and how to create it (573 words)
- [Imbue | Substack](/content/ideas/t/essays/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (158 words)
- [Imbue | Substack](/content/ideas/t/events/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (46 words)
- [From lawless spaces to true liberty: rethinking AI's role in society](/content/ideas/p/matt-boulos/index.html): Who will actually hold power in the age of intelligent machines? (33,380 words)
- [There will be a scientific theory of deep learning](/content/ideas/p/learning-mechanics/index.html): Introducing learning mechanics with Jamie Simon and Daniel Kunin (UC Berkeley) (29,265 words)
- [Attention as an Art Form - by Ashley Zhang and Adam Robbert](/content/ideas/p/attention-as-an-art-form/index.html): Watch now | Art of Being Human #2 with Adam Robbert (19,228 words)
- [From bare metal to a 70B model: infrastructure set-up and scripts - Imbue](/content/blog/70b-infrastructure/index.html) (6,528 words)
- [A PyTorch implementation of SIMONe - Imbue](/content/blog/2022-04-14-simone/index.html) (2,883 words)
- [Imbue Blog: Writings about honest (and punk!) software - Imbue](/content/blog/index.html): Product updates, research, and field notes on building software at the edge of AI technologies.From the team behind Sculptor, mngr, Blueprint, Vet, and more. (1,646 words)
- [Understanding self-supervised and contrastive learning with "Bootstrap Your Own Latent" (BYOL) - Imbue](/content/blog/2020-08-24-understanding-self-supervised-contrastive-learning.html) (2,648 words)
- [VICReg: Tutorial and lightweight PyTorch implementation - Imbue](/content/blog/2022-04-21-vicreg/index.html) (1,847 words)
- [A case study in testing with 100+ of Claude agents in parallel - Imbue](/content/blog/mngr_part_2/index.html): How we're using `mngr` to create (and improve) its own tests and documentation. (2,191 words)
- [Scaling Laws For Every Hyperparameter Via Cost-Aware HPO - Imbue](/content/blog/carbs/index.html) (1,764 words)
- [Usefully run hundreds of Claudes in parallel with mngr - Imbue](/content/blog/mngr/index.html): How to use `mngr` to massively scale the creation of high-quality software. (1,840 words)
- [Imbue Website Terms of Service - Imbue](/content/website/terms/index.html): The terms that govern your use of imbue.com. Read our rules on access, content, intellectual property, and acceptable use of the site. (1,696 words)
- [Jack Parker-Holder, DeepMind: On open-endedness, evolving agents and environments, online adaptation, and offline learning - Imbue](/content/blog/2022-12-06-podcast-episode-24-jack-parker-holder/index.html) (605 words)
- [Fine-grained HTTP permissions for AI agents - Imbue](/content/blog/fine-grained-http-permissions-for-ai-agents/index.html): How to lock down what AI agents can do with APIs they have access to. Fine-grained permissions for Latchkey using JSON schemas. (867 words)
- [Malleable software and human agency - Imbue](/content/ideas/p/geoffrey-litt/index.html): A conversation with Geoffrey Litt, design engineer at Notion, on shaping software like clay (34,173 words)
- [The state of LLM ethical decision-making - Imbue](/content/blog/llm-ethics/index.html) (721 words)
- [Noam Brown, FAIR: On achieving human-level performance in poker and Diplomacy, and the power of spending compute at inference time - Imbue](/content/blog/2023-02-09-podcast-episode-27-noam-brown/index.html) (557 words)
- [Ari Morcos, DatologyAI: On leveraging data to democratize model training - Imbue](/content/blog/2024-07-11-podcast-episode-36-ari-morcos/index.html) (1,078 words)
- [Empowering humans in the age of AI - Imbue](/content/blog/empowering-humans-in-the-age-of-ai/index.html): We’re making software creation accessible to all, so every person can have the power to shape their computational future. (1,175 words)
- [Sugandha Sharma, MIT: On biologically inspired neural architectures, how memories can be implemented, and control theory - Imbue](/content/blog/2023-01-11-podcast-episode-26-sugandha-sharma/index.html) (640 words)
- [Imbue raises $200M to build AI systems that can reason and code - Imbue](/content/blog/introducing-imbue/index.html) (1,645 words)
- [Sergey Levine, UC Berkeley: On the bottlenecks to generalization, why simulation is doomed to succeed, and picking good research problems - Imbue](/content/blog/2023-03-01-podcast-episode-28-sergey-levine/index.html) (881 words)
- [Julian Chibane, MPI-INF: On 3D reconstruction using implicit functions - Imbue](/content/blog/2021-03-05-podcast-episode-6-julian-chibane/index.html) (784 words)
- [Martin Arjovsky, INRIA: On benchmarks for robustness and geometric information theory - Imbue](/content/blog/2021-10-14-podcast-episode-15-martin-arjovsky/index.html) (458 words)
- [Jonathan Frankle, MIT: On the lottery ticket hypothesis and the science of deep learning - Imbue](/content/blog/2021-09-09-podcast-episode-13-jonathan-frankle/index.html) (832 words)
- [Introducing Generally Intelligent - Imbue](/content/blog/launch/index.html) (1,038 words)
- [Work with us! - Imbue](/content/careers/index.html): We look for people who care deeply about reinventing computing, and who enjoy questioning assumptions, building from first principles, and operating with high agency. If that sounds like you, read on. (420 words)
- [Hattie Zhou, Mila: On supermasks, iterative learning, and fortuitous forgetting - Imbue](/content/blog/2022-10-14-podcast-episode-20-hattie-zhou/index.html) (506 words)
- [Our approach to safety - Imbue](/content/blog/safety/index.html) (347 words)
- [There Will Be a Scientific Theory of Deep Learning - Imbue](/content/blog/2026-04-24-podcast-episode-38-jamie-simon-and-daniel-kunin.html) (779 words)
- [Closed vs. open AI ecosystems: Kanjun on Decoder - Imbue](/content/blog/closed-vs-open-ai-ecosystems-kanjun-on-decoder/index.html) (340 words)
- [The AI paradox: Kanjun at the 2024 Progress Conference - Imbue](/content/blog/the-ai-paradox/index.html) (100 words)
- [Developing agents that can reason and code: Kanjun and Josh on the No Priors podcast - Imbue](/content/blog/no-priors/index.html) (190 words)
- [Human agency vs. “agentic” AI: Kanjun on the NonZero podcast - Imbue](/content/blog/human-agency-vs-agentic-ai-kanjun-on-the-nonzero-podcast.html) (106 words)
- [The future of AI is human-centered: Kanjun on the Pioneers of AI podcast - Imbue](/content/blog/pioneers-of-ai/index.html) (109 words)
- [Joel Lehman, OpenAI: On evolving intelligence, open-endedness, and reinforcement learning - Imbue](/content/blog/2021-02-16-podcast-episode-4-joel-lehman/index.html) (382 words)
- [A more empowering view of general agents: Kanjun on NYSE TV - Imbue](/content/blog/a-more-empowering-view-of-general-agents-kanjun-on-nyse-tv.html) (18 words)
- [Agency in the age of AI: Kanjun on the Minus One podcast - Imbue](/content/blog/agency-in-the-age-of-ai/index.html) (61 words)
- [A Healthy Ecosystem for AI Agents: Introducing the AFI - Imbue](/content/blog/a-healthy-ecosystem-for-ai-agents-introducing-the-afi.html) (254 words)
- [LLM-based Evolution as a Universal Optimizer - Imbue](/content/blog/2026-02-27-darwinian-evolver/index.html) (2,110 words)
- [Open-sourcing CARBS: how we used our hyperparameter optimizer to scale up to a 70B-parameter language model - Imbue](/content/blog/70b-carbs/index.html) (3,855 words)
- [Sarah Jane Hong, Latent Space: On neural rendering & research process - Imbue](/content/blog/2021-01-06-podcast-episode-2-sarah/index.html) (6,538 words)
- [Beating ARC-AGI-2 with Code Evolution - Imbue](/content/blog/2026-02-27-arc-agi-2-evolution/index.html) (2,625 words)
- [Imbue Product Terms of Use - Imbue](/content/terms/index.html): The terms that govern your use of Imbue products and the research preview platform. Read our rules on access, accounts, and acceptable use. (3,786 words)
- [Glenn McDonald: On Dungeons & Dragons, data alchemy, and radically open software - Imbue](/content/blog/2025-01-30-glenn-team-spotlight/index.html) (1,289 words)
- [Appendix for "Understanding self-supervised and contrastive learning with 'Bootstrap Your Own Latent' (BYOL)" - Imbue](/content/blog/2020-08-24-appendix-for-understanding-self-supervised-contrastive-learning.html) (1,371 words)
- [Imbue Website Privacy Policy - Imbue](/content/website/privacy/index.html): How Imbue collects, uses, and shares personal information when you visit imbue.com. Learn what we collect, why, and the privacy choices you can make. (2,738 words)
- [Where are the good policy ideas? Analyzing the NTIA's AI RFC - Imbue](/content/blog/ntia-rfc-analysis/index.html) (2,112 words)
- [Bryden Fogelman: On coding agents and finding meaningful work at Imbue - Imbue](/content/blog/2023-09-06-bryden-team-spotlight/index.html) (1,527 words)
- [Keystone-Eval: Our agentic benchmark for self-configuring agents - Imbue](/content/blog/keystone-eval/index.html): How we built a reusable agentic benchmark and used it to optimize Keystone (2,587 words)
- [How Offload Works: Inside the Rust CLI that sped up our tests by 6x - Imbue](/content/blog/offload-how-it-works/index.html): A technical deep-dive into Offload's architecture (1,657 words)
- [Offload: We made agents run our test suite 6x faster - Imbue](/content/blog/offload/index.html): Here's how you can do it too (1,472 words)
- [Every mngr plugin, explained - Imbue](/content/blog/every-mngr-plugin-explained/index.html): A field guide to mngr's plugins: spin up agents on Modal or a VPS, schedule nightly runs, pair with a remote Claude in real time, map-reduce your test suite. (2,108 words)
- [How AI code review can make correct code worse - Imbue](/content/blog/2026-04-29-how-ai-code-review-can-make-correct-code-worse.html): Lessons from hundreds of trials on coding benchmarks (1,448 words)
- [Drew Linsley, Brown: On inductive biases for vision and generalization - Imbue](/content/blog/2021-04-01-podcast-episode-9-drew-linsley/index.html) (1,606 words)
- [Methodology for analyzing the individual comments to the NTIA’s AI RFC - Imbue](/content/blog/ntia-rfc-analysis-individual-methodology/index.html) (2,854 words)
- [Dylan Hadfield-Menell, UC Berkeley/MIT: On the value alignment problem in AI - Imbue](/content/blog/2021-05-11-podcast-episode-10-dylan-hadfield-menell.html) (1,031 words)
- [Training a 70B model from scratch: open-source tools, evaluation datasets, and learnings - Imbue](/content/blog/70b-intro/index.html) (823 words)
- [A more radical Imbue - Imbue](/content/blog/a-more-radical-imbue/index.html) (1,505 words)
- [The need for trust: Analyzing the individual comments to the NTIA’s AI RFC - Imbue](/content/blog/ntia-rfc-analysis-individual/index.html) (1,651 words)
- [Yujia Huang, Caltech: On neuro-inspired generative models - Imbue](/content/blog/2021-03-18-podcast-episode-7-yujia-huang/index.html) (1,007 words)
- [Towards guardrails, not guidelines: a policy framework for powerful AI systems - Imbue](/content/blog/common-good/index.html) (2,230 words)
- [Blueprint: One-shot bigger coding tasks - Imbue](/content/blog/blueprint/index.html): A planning copilot that asks smart questions and helps one-shot complex tasks with confidence (959 words)
- [Seth Lazar, Australian National University: On legitimate power, moral nuance, and the political philosophy of AI - Imbue](/content/blog/2024-03-12-podcast-episode-34-seth-lazar/index.html) (1,056 words)
- [On the Stepwise Nature of Self-Supervised Learning - Imbue](/content/blog/ssl_stepwise/index.html) (1,528 words)
- [Yash Sharma, MPI-IS: On generalizability, causality, and disentanglement - Imbue](/content/blog/2021-09-24-podcast-episode-14-yash-sharma/index.html) (566 words)
- [Imbue Product Privacy Policy - Imbue](/content/privacy/index.html): How Imbue handles personal information across our products and research preview. Learn what we collect, how we use it, and your privacy rights and choices. (2,262 words)
- [Choices and Knives - Imbue](/content/blog/choices-and-knives/index.html) (1,609 words)
- [Chelsea Finn, Stanford: On the biggest bottlenecks in robotics and reinforcement learning - Imbue](/content/blog/2022-11-03-podcast-episode-21-chelsea-finn/index.html) (692 words)
- [Rylan Schaeffer, Stanford: On investigating emergent abilities and challenging dominant research ideas - Imbue](/content/blog/2024-09-19-podcast-episode-37-rylan-schaeffer/index.html) (868 words)
- [How to usefully run 1,000 agents in parallel - Imbue](/content/blog/how-to-usefully-run-1-000-agents-in-parallel/index.html) (171 words)
- [Tri Dao, Stanford: On FlashAttention and sparsity, quantization, and efficient inference - Imbue](/content/blog/2024-02-08-podcast-episode-33-tri-dao/index.html) (656 words)
- [Sculptor: the missing UI for parallel coding agents - Imbue](/content/blog/sculptor-announce/index.html): We rebuilt Sculptor from the ground up, thanks to your feedback. Now you can run multiple coding agents in safe containers and instantly see their changes in action. (790 words)
- [Imbue Community: Where builders share with each other - Imbue](/content/community/index.html): Join a community of builders shaping open AI. Get early access, share feedback, and meet us on X, Substack, and at events in Hayes Valley, San Francisco. (169 words)
- [Celeste Kidd, UC Berkeley: On attention and curiosity, how we form beliefs, and where certainty comes from - Imbue](/content/blog/2022-11-22-podcast-episode-23-celeste-kidd/index.html) (565 words)
- [Jim Fan, NVIDIA: On foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant - Imbue](/content/blog/2023-03-08-podcast-episode-29-jim-fan/index.html) (834 words)
- [A PyTorch implementation of slot attention - Imbue](/content/blog/2021-03-09-slot-attention/index.html) (574 words)
- [Oleh Rybkin, UPenn: On exploration and planning with world models - Imbue](/content/blog/2022-07-11-podcast-episode-18-oleh-rybkin/index.html) (586 words)
- [Percy Liang, Stanford: On the paradigm shift and societal effects of foundation models - Imbue](/content/blog/2024-05-09-podcast-episode-35-percy-liang/index.html) (1,420 words)
- [Bouncer: Heal your feed - Imbue](/content/blog/bouncer/index.html): Twitter's algorithm is optimized for attention, not for you. Bouncer changes that. (570 words)
- [Vincent Sitzmann, MIT: On neural scene representations for computer vision and more general AI - Imbue](/content/blog/2021-05-19-podcast-episode-11-vincent-sitzmann/index.html) (975 words)
- [What's next in AI?: Kanjun and Reid Hoffman at Config 2023 - Imbue](/content/blog/whats-next/index.html) (740 words)
- [Why AI agents don’t work (yet): Kanjun on the Latent Space podcast - Imbue](/content/blog/latent-space-imbue/index.html) (166 words)
- [One-person enterprise: Kanjun at the World Economic Forum Annual Meeting - Imbue](/content/blog/world-economic-forum-annual-meeting/index.html): As AI evolves from a support tool to a co-pilot, the dynamics of building and running a company are changing. Companies are increasingly relying on technology enabling one individual to achieve what previously required entire teams. (93 words)
- [Nicklas Hansen, UCSD: On long-horizon planning and why algorithms don't drive research progress - Imbue](/content/blog/2022-12-16-podcast-episode-25-nicklas-hansen/index.html) (374 words)
- ['We need legal protections from AI’s risks': Matt Boulos in The Hill - Imbue](/content/blog/legal-protections-op-ed/index.html) (814 words)
- [How we made sandboxed coding agents 10x faster to start - Imbue](/content/blog/containers/index.html) (1,168 words)
- [Our technical hiring process - Imbue](/content/blog/2023-09-06-our-technical-hiring-process/index.html) (1,290 words)
- [Jamie Simon, UC Berkeley: On theoretical principles for how neural networks learn and generalize - Imbue](/content/blog/2023-06-29-podcast-episode-32-jamie-simon/index.html) (556 words)
- [Jacob Steinhardt, UC Berkeley: On machine learning safety, alignment and measurement - Imbue](/content/blog/2021-06-17-podcast-episode-12-jacob-steinhardt/index.html) (728 words)
- [Andrew Lampinen, DeepMind: On symbolic behavior, mental time travel, and insights from psychology - Imbue](/content/blog/2022-02-28-podcast-episode-17-andrew-lampinen/index.html) (692 words)
- [Minqi Jiang, UCL: On environment and curriculum design for general RL agents - Imbue](/content/blog/2022-07-19-podcast-episode-19-minqi-jiang/index.html) (431 words)
- [Giancarlo Kerg, Mila: On approaching deep learning from mathematical foundations - Imbue](/content/blog/2021-03-26-podcast-episode-8-giancarlo-kerg/index.html) (1,139 words)
- [Imbue Raises Fresh Capital, Amazon's Alexa Fund Joins Investor Lineup - Imbue](/content/blog/imbue-raises-fresh-capital/index.html) (338 words)
- [Transforming AI agents into personal collaborators: Kanjun on the NVIDIA AI Podcast - Imbue](/content/blog/transforming-ai-agents-into-personal-collaborators/index.html) (133 words)
- [Yilun Du, MIT: On energy-based models, implicit functions, and modularity - Imbue](/content/blog/2021-12-21-podcast-episode-16-yilun-du/index.html) (367 words)
- [Avalon: A Benchmark for RL Generalization Using Procedurally Generated Worlds (NeurIPS Spotlight Paper) - Imbue](/content/blog/avalon/index.html) (593 words)
- [Our submission to the NTIA's AI accountability RFC - Imbue](/content/blog/ntia-rfc/index.html) (1,546 words)
- [Reinventing AI agents: Kanjun on the Gradient Dissent podcast - Imbue](/content/blog/reinventing-ai-agents/index.html) (134 words)
- [Latchkey: Use plain old REST instead of MCP - Imbue](/content/blog/latchkey/index.html) (496 words)
- [Archit Sharma, Stanford: On unsupervised and autonomous reinforcement learning - Imbue](/content/blog/2022-11-17-podcast-episode-22-archit-sharma/index.html) (537 words)
- [Update: The state of LLM ethical decision-making - Imbue](/content/blog/llm-ethics-update/index.html) (489 words)
- [Training >70B LLMs on 10,000 H100 clusters: Josh on the Latent Space podcast - Imbue](/content/blog/training-greater-than-70b-llms-on-10000-h100-clusters.html) (383 words)
- [Jupyter Ascending: Sync notebooks to your editor - Imbue](/content/blog/2021-10-14-jupyter-ascending/index.html) (324 words)
- [Ben Eysenbach, CMU: On designing simpler and more principled RL algorithms - Imbue](/content/blog/2023-03-22-podcast-episode-30-ben-eysenbach/index.html) (441 words)
- [Use your unused Claude tokens for good - Imbue](/content/blog/unused-claude-tokens-for-good/index.html): mngr makes it easy to use your Claude quota before it times out! (561 words)
- [Bill Thompson, UC Berkeley: On how cultural evolution shapes knowledge acquisition - Imbue](/content/blog/2023-03-29-podcast-episode-31-bill-thompson/index.html) (425 words)
- [The ethical and practical future of AI: Matt Boulos at the MIT Club of Northern California - Imbue](/content/blog/the-ethical-and-practical-future-of-ai/index.html) (35 words)
- [Vet: Prevent coding agents from making mistakes - Imbue](/content/blog/vet/index.html) (247 words)
- [Introducing Keystone: Building self-configuring agents - Imbue](/content/blog/keystone/index.html) (462 words)
- [Katja Schwarz, MPI-IS: On GANs, implicit functions, and 3D scene understanding - Imbue](/content/blog/2021-02-23-podcast-episode-5-katja-schwarz/index.html) (322 words)
- [Sculptor: Catch and fix issues as you code - Imbue](/content/blog/sculptor/index.html): Sculptor is a coding agent environment that applies engineering discipline to catch issues, write tests, and improve your code—all while you work in your favorite editor. (522 words)
- [How to build robust software systems with AI: Josh at the AI Engineer World's Fair 2025 - Imbue](/content/blog/how-to-build-robust-software-systems-with-ai-josh-at-the-ai-engineer-world-s-fair-2025.html) (51 words)
- [Imbue Products: Honest (and punk!) software - Imbue](/content/products/index.html): Open developer tools for AI coding agents. Run agents in parallel, catch their mistakes, and own your workspace with Sculptor, mngr, Vet, and more. (95 words)
- [Keeping humans in the loop: Josh at the Democratize Intelligence Summit - Imbue](/content/blog/keeping-humans-in-the-loop-josh-at-the-democratize-intelligence-summit.html) (190 words)
- [Generally Intelligent #0: A podcast for deep learning researchers - Imbue](/content/blog/2020-12-06-podcast-episode-0-intro/index.html) (380 words)
- [Digital Freedom Depends on Access Rights: Matt Boulos in Lawfare - Imbue](/content/blog/digital-freedom-depends-on-access-rights-matt-boulos-in-lawfare.html) (244 words)
- [Cinjon Resnick, NYU: On activity and scene understanding - Imbue](/content/blog/2021-01-31-podcast-episode-3-cinjon/index.html) (11,533 words)

## Products

- [Offload: Parallelize tests across 200+ cloud sandboxes in one command - Imbue](/content/product/offload/index.html): Offload is a Rust CLI your agents invoke directly. It spreads your existing test suite across 200+ isolated cloud sandboxes via Modal. One TOML config, no test rewrites.  (675 words)
- [Bouncer: Filter (and heal) your feed - Imbue](/content/product/bouncer/index.html): Bouncer is an app and browser extension that filters out what you don’t want to see on X: a phrase, a topic, a tone. And because X's algorithm learns from what you don’t look at, Bouncer heals your feed. (513 words)
- [Keystone: Teach your repo how to run itself - Imbue](/content/product/keystone/index.html): Keystone self-configures a working devcontainer for any git repo, all on its own. Give it a repo. Get back a Dockerfile, devcontainer.json, and a passing test runner. It runs a coding agent inside a sandboxed Modal environment so your machine is never touched. (599 words)
- [Sculptor: The extensible, parallel coding workspace you own. - Imbue](/content/product/sculptor/index.html): Sculptor is an extensible, open source app for running coding agents in parallel. (808 words)
- [Latchkey: Credential layer for local AI agents - Imbue](/content/product/latchkey/index.html): A command-line tool that injects credentials to curl requests to known public APIs. (465 words)
- [mngr: Run any coding agent in parallel, anywhere - Imbue](/content/product/mngr/index.html): mngr is a Unix-style CLI tool for managing coding agents. Seamlessly scale from a single local Claude to 100s of agents across remote hosts, containers, and sandboxes. (465 words)
- [Vet: Catch your coding agent’s mistakes - Imbue](/content/product/vet/index.html): Vet is a fast and local code review tool open-sourced by the Imbue team. It’s concise where others are verbose. (420 words)
- [Blueprint: One-shot bigger coding tasks - Imbue](/content/product/blueprint/index.html): Blueprint reads your codebase, interviews you, then writes a plan a coding agent can one-shot. Available as VS Code, Cursor, Windsurf extensions plus a harness-agnostic agent skill. (469 words)

## About Pages

- [About - Imbue](/content/ideas/about/index.html): Ideas on the future we want, and how to build the technologies and systems that get us there. Click to read Imbue, a Substack publication with thousands of subscribers. (267 words)
- [We build tools that put software under your control - Imbue](/content/about/index.html): Today, the software and agents in our lives don’t work for us—they work for the company that built them. Closed platforms lock you into their ecosystems, limit what you can do, and serve their incentives over yours.  We're building the alternative: an open agent ecosystem where you can understand, modify, and verify your agents, so they work only in your interest. (416 words)

## Resources

- [Full Page Index](/index.html): Browse all cached pages with rich metadata
- [About This Cache](/content/about.html): Methodology, technical details, and usage guidelines
- [XML Sitemap](/sitemap.xml): Machine-readable sitemap for crawler discovery
- [Robots.txt](/robots.txt): Crawler directives