# Blog

## Filters

- [Use your unused Claude tokens for good](/content/blog/unused-claude-tokens-for-good/index.html)
- [Every mngr plugin, explained](/content/blog/every-mngr-plugin-explained/index.html)
- [Fine-grained HTTP permissions for AI agents](/content/blog/fine-grained-http-permissions-for-ai-agents/index.html)
- [How AI code review can make correct code worse](/content/blog/2026-04-29-how-ai-code-review-can-make-correct-code-worse/index.html)
- [Blueprint: One-shot bigger coding tasks](/content/blog/blueprint/index.html)
- [There Will Be a Scientific Theory of Deep Learning](/content/blog/2026-04-24-podcast-episode-38-jamie-simon-and-daniel-kunin/index.html)
- [Keystone-Eval: Our agentic benchmark for self-configuring agents](/content/blog/keystone-eval/index.html)
- [How to usefully run 1,000 agents in parallel](/content/blog/how-to-usefully-run-1-000-agents-in-parallel/index.html)
- [Bouncer: Heal your feed](/content/blog/bouncer/index.html)
- [How Offload Works: Inside the Rust CLI that sped up our tests by 6x](/content/blog/offload-how-it-works/index.html)
- [A case study in testing with 100+ of Claude agents in parallel](/content/blog/mngr_part_2/index.html)
- [Usefully run hundreds of Claudes in parallel with mngr](/content/blog/mngr/index.html)
- [Introducing Keystone: Building self-configuring agents](/content/blog/keystone/index.html)
- [Offload: We made agents run our test suite 6x faster](/content/blog/offload/index.html)
- [Vet: Prevent coding agents from making mistakes](/content/blog/vet/index.html)
- [Latchkey: Use plain old REST instead of MCP](/content/blog/latchkey/index.html)
- [LLM-based Evolution as a Universal Optimizer](/content/blog/2026-02-27-darwinian-evolver/index.html)
- [Beating ARC-AGI-2 with Code Evolution](/content/blog/2026-02-27-arc-agi-2-evolution/index.html)
- [A more radical Imbue](/content/blog/a-more-radical-imbue/index.html)
- [Choices and Knives](/content/blog/choices-and-knives/index.html)
- [Digital Freedom Depends on Access Rights: Matt Boulos in Lawfare](/content/blog/digital-freedom-depends-on-access-rights-matt-boulos-in-lawfare/index.html)
- [How we made sandboxed coding agents 10x faster to start](/content/blog/containers/index.html)
- [A Healthy Ecosystem for AI Agents: Introducing the AFI](/content/blog/a-healthy-ecosystem-for-ai-agents-introducing-the-afi/index.html)
- [Closed vs. open AI ecosystems: Kanjun on Decoder](/content/blog/closed-vs-open-ai-ecosystems-kanjun-on-decoder/index.html)
- [Sculptor: the missing UI for parallel coding agents](/content/blog/sculptor-announce/index.html)
- [How to build robust software systems with AI: Josh at the AI Engineer World's Fair 2025](/content/blog/how-to-build-robust-software-systems-with-ai-josh-at-the-ai-engineer-world-s-fair-2025/index.html)
- [Beyond the Prototype: Using AI to Write High-Quality Code](https://www.youtube.com/watch?v=x_1EumTaXeE)
- [Empowering humans in the age of AI](/content/blog/empowering-humans-in-the-age-of-ai/index.html)
- [Sculptor: Catch and fix issues as you code](/content/blog/sculptor/index.html)
- [One-person enterprise: Kanjun at the World Economic Forum Annual Meeting](/content/blog/world-economic-forum-annual-meeting/index.html)
- [The future of AI is human-centered: Kanjun on the Pioneers of AI podcast](/content/blog/pioneers-of-ai/index.html)
- [Glenn McDonald: On Dungeons & Dragons, data alchemy, and radically open software](/content/blog/2025-01-30-glenn-team-spotlight/index.html)
- [Keeping humans in the loop: Josh at the Democratize Intelligence Summit](/content/blog/keeping-humans-in-the-loop-josh-at-the-democratize-intelligence-summit/index.html)
- [Human agency vs. “agentic” AI: Kanjun on the NonZero podcast](/content/blog/human-agency-vs-agentic-ai-kanjun-on-the-nonzero-podcast/index.html)
- [A more empowering view of general agents: Kanjun on NYSE TV](/content/blog/a-more-empowering-view-of-general-agents-kanjun-on-nyse-tv/index.html)
- [Transforming AI agents into personal collaborators: Kanjun on the NVIDIA AI Podcast](/content/blog/transforming-ai-agents-into-personal-collaborators/index.html)
- [The ethical and practical future of AI: Matt Boulos at the MIT Club of Northern California](/content/blog/the-ethical-and-practical-future-of-ai/index.html)
- [The AI paradox: Kanjun at the 2024 Progress Conference](/content/blog/the-ai-paradox/index.html)
- [Rylan Schaeffer, Stanford: On investigating emergent abilities and challenging dominant research ideas](/content/blog/2024-09-19-podcast-episode-37-rylan-schaeffer/index.html)
- [Agency in the age of AI: Kanjun on the Minus One podcast](/content/blog/agency-in-the-age-of-ai/index.html)
- [“The AI startup Imbue is building a service that will let you DIY a bot for any task, large or small,” said Kanjun Qiu, its chief executive and co-founder. Rather than relying on companies to build bots for us, Qiu said Imbue's service aims to give people more control in a bot-powered world.](https://www.wsj.com/tech/ai/ai-agents-autonomous-bots-human-tasks-913c0ce5)
- [Reinventing AI agents: Kanjun on the Gradient Dissent podcast](/content/blog/reinventing-ai-agents/index.html)
- [These Are the 10 AI Startups to Watch in 2024](https://www.bloomberg.com/features/2024-top-ai-startups/)
- [Ari Morcos, DatologyAI: On leveraging data to democratize model training](/content/blog/2024-07-11-podcast-episode-36-ari-morcos/index.html)
- [Kanjun Qiu, CEO and founder of the AI startup Imbue, which is working on agents that can reason and code, likens the state of agents to where self-driving cars were just over a decade ago.](https://www.technologyreview.com/2024/07/05/1094711/what-are-ai-agents/)
- [From bare metal to a 70B model: infrastructure set-up and scripts](/content/blog/70b-infrastructure/index.html)
- [Training >70B LLMs on 10,000 H100 clusters: Josh on the Latent Space podcast](/content/blog/training-greater-than-70b-llms-on-10000-h100-clusters/index.html)
- [Open-sourcing CARBS: how we used our hyperparameter optimizer to scale up to a 70B-parameter language model](/content/blog/70b-carbs/index.html)
- [Training a 70B model from scratch: open-source tools, evaluation datasets, and learnings](/content/blog/70b-intro/index.html)
- [Sanitized open-source datasets for natural language and code understanding: how we evaluated our 70B model](/content/blog/70b-evals/index.html)
- [“The real potential of agents, if we do a good job with invention, is that we can give individual people much more power over our computing environments.” — Kanjun Qiu, CEO](https://app.frame.io/reviews/2f0cc6a2-8da4-44aa-80be-b2c2a098eb61/509a74fc-b0ca-46d5-b08c-5e72d86dbf85)
- [“The lack of openness has definitely been noticeable over the past year and it's put a damper on academic investigations,” — Josh Albrecht, CTO](https://www.semafor.com/article/02/16/2024/tech-companies-go-dark-about-ai-advances)
- [Percy Liang, Stanford: On the paradigm shift and societal effects of foundation models](/content/blog/2024-05-09-podcast-episode-35-percy-liang/index.html)
- ['We need legal protections from AI’s risks': Matt Boulos in The Hill](/content/blog/legal-protections-op-ed/index.html)
- [I think of it as really being able to [help us] control information environments… once we have control over information environments, we’ll feel a lot more empowered,” Qiu said. “Every single person’s vision can come to life.](https://blogs.nvidia.com/blog/chat-catanzaro-qiu/)
- [Technology that we build reflects the values of its creators,” says Qiu. “There’s a much more human way to think about building out this technology, in a way that’s empowering to people.](https://fortune.com/2024/03/12/smart-strategies-ai-investing-diversity/)
- [Seth Lazar, Australian National University: On legitimate power, moral nuance, and the political philosophy of AI](/content/blog/2024-03-12-podcast-episode-34-seth-lazar/index.html)
- [“What I’m most excited about is that I think [AI] is one of the biggest things that can unlock our potential.” — Kanjun Qiu, CEO](https://abcnews.go.com/Technology/video/rise-ai-job-107545879)
- [Imbue joins Department of Commerce’s AI Safety Institute Consortium](https://www.commerce.gov/news/press-releases/2024/02/biden-harris-administration-announces-first-ever-consortium-dedicated)
- [Tri Dao, Stanford: On FlashAttention and sparsity, quantization, and efficient inference](/content/blog/2024-02-08-podcast-episode-33-tri-dao/index.html)
- [The future of AI is human-centered](https://www.youtube.com/watch?v=503RsthkV6M)
- [Imbue and Dell designed the custom system... to support rapid experimentation in developing AI systems](https://www.bloomberg.com/news/articles/2023-11-28/dell-inks-150-million-computing-deal-with-ai-startup-imbue)
- [We've partnered with Dell to build a cluster of 10,000 Nvidia H100 GPUs to train our models](https://www.prnewswire.com/news-releases/imbue-to-develop-next-generation-ai-models-with-150-million-dell-high-performance-computing-system-301998808.html#:~:text=Dell%20Technologies%20(NYSE%3A%20DELL),foundation%20models%20optimized%20for%20reasoning)
- [Developing agents that can reason and code: Kanjun and Josh on the No Priors podcast](/content/blog/no-priors/index.html)
- [Methodology for analyzing the individual comments to the NTIA’s AI RFC](/content/blog/ntia-rfc-analysis-individual-methodology/index.html)
- [The need for trust: Analyzing the individual comments to the NTIA’s AI RFC](/content/blog/ntia-rfc-analysis-individual/index.html)
- [“We want to enable agents that people can actually use and that work reliably, and ultimately the vision is to reinvent the personal computer so that it’s actually truly personal.” Kanjun Qiu](https://www.reuters.com/markets/deals/ai-startup-imbue-gets-12-million-follow-on-funding-2023-10-19/)
- [Imbue Raises Fresh Capital, Amazon's Alexa Fund Joins Investor Lineup](/content/blog/imbue-raises-fresh-capital/index.html)
- [Why AI agents don’t work (yet): Kanjun on the Latent Space podcast](/content/blog/latent-space-imbue/index.html)
- [Imbue raises $200M to build AI systems that can reason and code](/content/blog/introducing-imbue/index.html)
- [Imbue raises $200 Million Series B](https://www.bloomberg.com/news/videos/2023-09-07/bloomberg-technology-09-07-2023)
- [Our technical hiring process](/content/blog/2023-09-06-our-technical-hiring-process/index.html)
- [Bryden Fogelman: On coding agents and finding meaningful work at Imbue](/content/blog/2023-09-06-bryden-team-spotlight/index.html)
- [In-person culture and habits](/content/blog/2023-09-01-in-person-culture-and-habits/index.html)
- [Where are the good policy ideas? Analyzing the NTIA's AI RFC](/content/blog/ntia-rfc-analysis/index.html)
- [“The real challenge is building systems with robust reasoning,” said Qiu.](https://www.reuters.com/technology/race-towards-autonomous-ai-agents-grips-silicon-valley-2023-07-17/)
- [What's next in AI?: Kanjun and Reid Hoffman at Config 2023](/content/blog/whats-next/index.html)
- [On the Stepwise Nature of Self-Supervised Learning](/content/blog/ssl_stepwise/index.html)
- [Jamie Simon, UC Berkeley: On theoretical principles for how neural networks learn and generalize](/content/blog/2023-06-29-podcast-episode-32-jamie-simon/index.html)
- [Scaling Laws For Every Hyperparameter Via Cost-Aware HPO](/content/blog/carbs/index.html)
- [Our submission to the NTIA's AI accountability RFC](/content/blog/ntia-rfc/index.html)
- [Towards guardrails, not guidelines: a policy framework for powerful AI systems](/content/blog/common-good/index.html)
- [Kanjun Qiu, the CEO of Generally Intelligent, an AI research company, has been buying advanced graphic chips since last year for her own servers, allowing her to ride out the current shortage.](https://www.wsj.com/articles/the-ai-boom-runs-on-chips-but-it-cant-get-enough-9f76f554)
- [“The biggest thing…that I find them to be helpful for is inspiring me as the user and helping me learn things that I wouldn’t have necessarily thought of on my own,” Josh Albrecht, CTO of Generally Intelligent](https://www.cnbc.com/2023/05/25/how-to-use-chatgpt-bard-other-chatbots.html)
- [Update: The state of LLM ethical decision-making](/content/blog/llm-ethics-update/index.html)
- [Bill Thompson, UC Berkeley: On how cultural evolution shapes knowledge acquisition](/content/blog/2023-03-29-podcast-episode-31-bill-thompson/index.html)
- [“Over the next few months and years, we can expect much of the internet to get connected to large language models.” — Kanjun Qiu, CEO](https://www.wired.com/story/chatgpt-plugins-openai/)
- [Instead of relying exclusively on outside providers, the AI startup Generally Intelligent has begun purchasing its own graphics chips to power its research.](https://www.wsj.com/articles/microsoft-google-amazon-look-to-generative-ai-to-lift-cloud-businesses-7159a43f/)
- [Ben Eysenbach, CMU: On designing simpler and more principled RL algorithms](/content/blog/2023-03-22-podcast-episode-30-ben-eysenbach/index.html)
- [Jim Fan, NVIDIA: On foundation models for embodied agents, scaling data, and why prompt engineering will become irrelevant](/content/blog/2023-03-08-podcast-episode-29-jim-fan/index.html)
- [Sergey Levine, UC Berkeley: On the bottlenecks to generalization, why simulation is doomed to succeed, and picking good research problems](/content/blog/2023-03-01-podcast-episode-28-sergey-levine/index.html)
- [The state of LLM ethical decision-making](/content/blog/llm-ethics/index.html)
- [Noam Brown, FAIR: On achieving human-level performance in poker and Diplomacy, and the power of spending compute at inference time](/content/blog/2023-02-09-podcast-episode-27-noam-brown/index.html)
- [Sugandha Sharma, MIT: On biologically inspired neural architectures, how memories can be implemented, and control theory](/content/blog/2023-01-11-podcast-episode-26-sugandha-sharma/index.html)
- [Nicklas Hansen, UCSD: On long-horizon planning and why algorithms don't drive research progress](/content/blog/2022-12-16-podcast-episode-25-nicklas-hansen/index.html)
- [Jack Parker-Holder, DeepMind: On open-endedness, evolving agents and environments, online adaptation, and offline learning](/content/blog/2022-12-06-podcast-episode-24-jack-parker-holder/index.html)
- [“But it really begs the question: what is understanding?” — Kanjun Qiu, CEO](https://time.com/6238781/chatbot-chatgpt-ai-interview/)
- [Celeste Kidd, UC Berkeley: On attention and curiosity, how we form beliefs, and where certainty comes from](/content/blog/2022-11-22-podcast-episode-23-celeste-kidd/index.html)
- [Archit Sharma, Stanford: On unsupervised and autonomous reinforcement learning](/content/blog/2022-11-17-podcast-episode-22-archit-sharma/index.html)
- [Chelsea Finn, Stanford: On the biggest bottlenecks in robotics and reinforcement learning](/content/blog/2022-11-03-podcast-episode-21-chelsea-finn/index.html)
- [Introducing Generally Intelligent](/content/blog/launch/index.html)
- [Our approach to safety](/content/blog/safety/index.html)
- [Avalon: A Benchmark for RL Generalization Using Procedurally Generated Worlds (NeurIPS Spotlight Paper)](/content/blog/avalon/index.html)
- [Among those contributing to the company's $20 million in initial funding (plus over $100 million in options) is Tom Brown, former engineering lead for OpenAI's GPT-3; former OpenAI robotics lead Jonas Schneider; Dropbox co-founders Drew Houston and Arash Ferdowsi; and the Astera Institute.](https://techcrunch.com/2022/10/20/generally-intelligent-secures-cash-from-openai-vets-to-build-capable-ai-systems/)
- [Hattie Zhou, Mila: On supermasks, iterative learning, and fortuitous forgetting](/content/blog/2022-10-14-podcast-episode-20-hattie-zhou/index.html)
- [Minqi Jiang, UCL: On environment and curriculum design for general RL agents](/content/blog/2022-07-19-podcast-episode-19-minqi-jiang/index.html)
- [Oleh Rybkin, UPenn: On exploration and planning with world models](/content/blog/2022-07-11-podcast-episode-18-oleh-rybkin/index.html)
- [VICReg: Tutorial and lightweight PyTorch implementation](/content/blog/2022-04-21-vicreg/index.html)
- [A PyTorch implementation of SIMONe](/content/blog/2022-04-14-simone/index.html)
- [Andrew Lampinen, DeepMind: On symbolic behavior, mental time travel, and insights from psychology](/content/blog/2022-02-28-podcast-episode-17-andrew-lampinen/index.html)
- [Yilun Du, MIT: On energy-based models, implicit functions, and modularity](/content/blog/2021-12-21-podcast-episode-16-yilun-du/index.html)
- [Martin Arjovsky, INRIA: On benchmarks for robustness and geometric information theory](/content/blog/2021-10-14-podcast-episode-15-martin-arjovsky/index.html)
- [Jupyter Ascending: Sync notebooks to your editor](/content/blog/2021-10-14-jupyter-ascending/index.html)
- [Yash Sharma, MPI-IS: On generalizability, causality, and disentanglement](/content/blog/2021-09-24-podcast-episode-14-yash-sharma/index.html)
- [Jonathan Frankle, MIT: On the lottery ticket hypothesis and the science of deep learning](/content/blog/2021-09-09-podcast-episode-13-jonathan-frankle/index.html)
- [Jacob Steinhardt, UC Berkeley: On machine learning safety, alignment and measurement](/content/blog/2021-06-17-podcast-episode-12-jacob-steinhardt/index.html)
- [Vincent Sitzmann, MIT: On neural scene representations for computer vision and more general AI](/content/blog/2021-05-19-podcast-episode-11-vincent-sitzmann/index.html)
- [Dylan Hadfield-Menell, UC Berkeley/MIT: On the value alignment problem in AI](/content/blog/2021-05-11-podcast-episode-10-dylan-hadfield-menell/index.html)
- [Drew Linsley, Brown: On inductive biases for vision and generalization](/content/blog/2021-04-01-podcast-episode-9-drew-linsley/index.html)
- [Giancarlo Kerg, Mila: On approaching deep learning from mathematical foundations](/content/blog/2021-03-26-podcast-episode-8-giancarlo-kerg/index.html)
- [Yujia Huang, Caltech: On neuro-inspired generative models](/content/blog/2021-03-18-podcast-episode-7-yujia-huang/index.html)
- [A PyTorch implementation of slot attention](/content/blog/2021-03-09-slot-attention/index.html)
- [Julian Chibane, MPI-INF: On 3D reconstruction using implicit functions](/content/blog/2021-03-05-podcast-episode-6-julian-chibane/index.html)
- [Katja Schwarz, MPI-IS: On GANs, implicit functions, and 3D scene understanding](/content/blog/2021-02-23-podcast-episode-5-katja-schwarz/index.html)
- [Joel Lehman, OpenAI: On evolving intelligence, open-endedness, and reinforcement learning](/content/blog/2021-02-16-podcast-episode-4-joel-lehman/index.html)
- [A new AI research lab: Josh on the Machine Learning Engineered podcast](/content/blog/2021-02-24/josh-albrecht-ml-engineered/index.html)
- [Cinjon Resnick, NYU: On activity and scene understanding](/content/blog/2021-01-31-podcast-episode-3-cinjon/index.html)
- [Sarah Jane Hong, Latent Space: On neural rendering & research process](/content/blog/2021-01-06-podcast-episode-2-sarah/index.html)
- [Kelvin Guu, Google AI: On language models & overlooked problems](/content/blog/2020-12-16-podcast-episode-1-kelvin/index.html)
- [Generally Intelligent #0: A podcast for deep learning researchers](/content/blog/2020-12-06-podcast-episode-0-intro/index.html)
- [Understanding self-supervised and contrastive learning with "Bootstrap Your Own Latent" (BYOL)](/content/blog/2020-08-24/understanding-self-supervised-contrastive-learning/index.html)
- [Appendix for "Understanding self-supervised and contrastive learning with 'Bootstrap Your Own Latent' (BYOL)"](/content/blog/2020-08-24-appendix-for-understanding-self-supervised-contrastive-learning/index.html)