#51 – Jeffrey Quesnelle on Nous Research, large language models, and the human mind

#52 – Michael Nielsen on being a wise optimist about science and technology

This is my conversation with Michael Nielsen, scientist, author, and research fellow at the Astera Institute. Timestamps: - (00:00:00) intro - (00:01:06) cultivating optimism amid existential risks - (00:07:16) asymmetric leverage - (00:12:09) are "unbiased" models even feasible? - (00:18:44) AI and the scientific method - (00:23:23) unlocking AI's full power through better interfaces - (00:30:33) sponsor: Splits - (00:31:18) AIs, independent agents or intelligent tools? - (00:35:47) autonomous military and weapons - (00:42:14) finding alignment - (00:48:28) aiming for specific moral outcomes with AI? - (00:54:42) freedom/progress vs safety - (00:57:46) provable beneficiary surveillance - (01:04:16) psychological costs - (01:12:40) the ingenuity gap Links: - Michael Nielsen: https://michaelnielsen.org/ - Michael Nielsen on X: https://x.com/michael_nielsen - Michael's essay on being a wise optimist about science and technology: https://michaelnotebook.com/optimism/ - Michael's Blog: https://michaelnotebook.com/ - The Ingenuity Gap (Tad Homer-Dixon): https://homerdixon.com/books/the-ingenuity-gap/ Thank you to our sponsor for making this podcast possible: - Splits: https://splits.org Into the Bytecode: - Sina Habibian on X: https://twitter.com/sinahab - Sina Habibian on Farcaster: https://warpcast.com/sinahab - Into the Bytecode: https://intothebytecode.com Disclaimer: This podcast is for informational purposes only. It is not financial advice nor a recommendation to buy or sell securities. The host and guests may hold positions in the projects discussed.

#51 – Jeffrey Quesnelle on Nous Research, large language models, and the human mind

This is my conversation with Jeffrey Quesnelle, cofounder of Nous Research. Timestamps: - (00:00:00) intro - (00:01:08) working with new technologies - (00:06:15) Nous Research origin story - (00:14:08) open frontiers in research - (00:26:07) fourier transforms for gradient compression - (00:32:58) math behind distributed training - (00:38:18) sponsor: Splits - (00:39:02) neural networks history and fundamentals - (00:51:29) the human mind and AI, hyperdimensional representation - (01:01:15) intuition and reasoning - (01:15:00) parallels with reinforcement learning - (01:19:15) the cat is out of the bag - (01:47:11) deeper mysteries Links: - Jeffrey Quesnelle: https://jeffq.com/ - Jeffrey Quesnelle on X: https://x.com/theemozilla - Nous Research: https://nousresearch.com/ - Psyche: https://nousresearch.com/nous-psyche/ Thank you to our sponsor for making this podcast possible: - Splits: https://splits.org Into the Bytecode: - Sina Habibian on X: https://twitter.com/sinahab - Sina Habibian on Farcaster: https://warpcast.com/sinahab - Into the Bytecode: https://intothebytecode.com Disclaimer: This podcast is for informational purposes only. It is not financial advice nor a recommendation to buy or sell securities. The host and guests may hold positions in the projects discussed.

#50 – Alexander Long on Pluralis Research and protocol learning for frontier models

This is my conversation with Alexander Long, Founder & CEO of Pluralis Research. Timestamps: - (00:00:00) intro - (00:00:55) collaborative training - (00:09:49) economics of training - (00:13:10) what is protocol learning? - (00:20:48) protocol learning design and politics - (00:33:39) sponsor: Splits - (00:34:22) hardware requirements - (00:41:53) adapting to the landscape - (00:49:53) open and closed models - (00:52:52) market structure with fully open models - (00:56:34) research and risks - (01:02:19) labor and national security - (01:10:58) looking to the future - (01:14:20) outro Links: - Alexander on X: https://x.com/_alexanderlong [https://x.com/_alexanderlong] - Alexander on Github: https://github.com/AlexanderJLong [https://github.com/AlexanderJLong] - Article 2: Protocol Learning, Protocol Models and the Great Convergence: https://www.pluralisresearch.com/p/article-2-protocol-learning-protocol [https://www.pluralisresearch.com/p/article-2-protocol-learning-protocol] - Decentralized Training Looms: https://www.pluralisresearch.com/p/decentralized-ai-looms [https://www.pluralisresearch.com/p/decentralized-ai-looms] Thank you to our sponsor for making this podcast possible: - Splits: https://splits.org [https://splits.org] Into the Bytecode: - Sina Habibian on X: https://twitter.com/sinahab - Sina Habibian on Farcaster - https://warpcast.com/sinahab [https://warpcast.com/sinahab] - Into the Bytecode: https://intothebytecode.com [https://intothebytecode.com] Disclaimer: this podcast is for informational purposes only. It is not financial advice nor a recommendation to buy or sell securities. The host and guests may hold positions in the projects discussed.

#49 – Jim Posen on cryptographic acceleration with Binius

This is my conversation with Jim Posen, the Cofounder and CTO at Irreducible. Timestamps: - (00:00:00) intro - (00:01:12) getting deeper into cryptography - (00:07:06) revisiting binary fields - (00:15:50) building the verifiable internet - (00:26:14) sponsor: Splits - (00:26:57) revival of binary fields - (00:38:55) Binius - (00:42:46) bringing Binius to production - (00:48:27) creating new hardware - (00:53:57) getting to v1 - (01:02:03) Ethereum stateless proofs and zkVMs - (01:13:21) outro Links: - Jim on X: https://x.com/jimpo_potamus [https://x.com/jimpo_potamus] - Jim on Github: https://github.com/jimpo [https://github.com/jimpo] - Irreducible: https://www.irreducible.com/ [https://www.irreducible.com/] - Binius: a Hardware-Optimized SNARK: https://www.irreducible.com/posts/binius-hardware-optimized-snark [https://www.irreducible.com/posts/binius-hardware-optimized-snark] - Binary Tower Fields are the Future of Verifiable Computing: https://www.irreducible.com/posts/binary-tower-fields-are-the-future-of-verifiable-computing [https://www.irreducible.com/posts/binary-tower-fields-are-the-future-of-verifiable-computing] Thank you to our sponsor for making this podcast possible: - Splits: https://splits.org [https://splits.org] Into the Bytecode: - Sina Habibian on X: https://twitter.com/sinahab [https://twitter.com/sinahab] - Sina Habibian on Farcaster - https://warpcast.com/sinahab [https://warpcast.com/sinahab] - Into the Bytecode: https://intothebytecode.com [https://intothebytecode.com] Disclaimer: this podcast is for informational purposes only. It is not financial advice nor a recommendation to buy or sell securities. The host and guests may hold positions in the projects discussed.

#48 – Michael Bentley on Euler, credit, and natural selection

This is my conversation with Michael Bentley, the cofounder and CEO of Euler and previously a dynamical systems postdoctoral research associate at Oxford. Timestamps: - (00:00:00) intro - (00:01:11) early DeFi and experiencing the 2008 crash - (00:06:52) interest rate design and PIDs - (00:20:40) risk management - (00:28:27) sponsor: Splits - (00:29:10) macro view of the world - (00:32:27) the fundamental importance of credit - (00:35:47) Euler protocol design - (00:39:08) modularity and integration - (00:54:40) network topology - (01:10:55) natural selection and dynamical systems - (01:26:33) outro Links: - Michael on X: https://x.com/euler_mab - Michael on Medium: https://medium.com/@euler-mab - Michael Bentley on the evolution of lending protocols: https://youtu.be/TW9Tv2Y0fkw - Euler: https://www.euler.finance Thank you to our sponsor for making this podcast possible: - Splits: https://splits.org Into the Bytecode: - Sina Habibian on X: https://twitter.com/sinahab - Sina Habibian on Farcaster - https://warpcast.com/sinahab - Into the Bytecode: https://intothebytecode.com  Disclaimer: this podcast is for informational purposes only. It is not financial advice nor a recommendation to buy or sell securities. The host and guests may hold positions in the projects discussed.