# duality (blogs) — RSS Amplifier Recent posts from the 2 feeds in the RSS Amplifier directory that cover duality. Page: Feed: --- ## [The Big Bang’s Smoking Gun: The Birth of the Elements](https://quantumworld.substack.com/p/the-big-bangs-smoking-gun-the-birth) _2026-07-29 · Marco Masi · Quantum Physics: An Overview of a Weird World_ The remarkable evidence hidden in the first atomic nuclei of the Universe ## [Quantum Mechanics and Consciousness](https://quantumworld.substack.com/p/quantum-mechanics-and-consciousness) _2026-07-16 · Marco Masi · Quantum Physics: An Overview of a Weird World_ Myth or Reality? ## [Quantum Biology](https://quantumworld.substack.com/p/quantum-biology) _2026-07-13 · Marco Masi · Quantum Physics: An Overview of a Weird World_ Myth or Reality? ## [The state vector and the Measurement Postulate- Book Excerpt #20](https://quantumworld.substack.com/p/the-state-vector-and-the-measurement-e3d) _2026-07-11 · Marco Masi · Quantum Physics: An Overview of a Weird World_ When Reality and Math Meet - Part II ## [Review and Mentoring of Independent Research](https://quantumworld.substack.com/p/review-and-mentoring-of-independent) _2026-06-27 · Marco Masi · Quantum Physics: An Overview of a Weird World_ Constructive feedback and mentoring for independent thinkers ## [ReplaySSM: Cache SSM Inputs, Not State](https://tridao.me/blog/2026/replayssm/) _2026-06-15 · Ze-Wei Liou · Tri Dao_ Current SSM decoding updates and writes recurrent state back to HBM every step. ReplaySSM instead caches recent inputs, reconstructing the state only when needed and otherwise computing the output directly from the cache. Figure 1a. ReplaySSM caches recent SSM inputs instead of storing the recurrent state every step, reconstructing states on the fly and writing back only when the buffer is full.… ## [Quantum Physics: PDF previews to explore the book](https://quantumworld.substack.com/p/quantum-physics-pdf-previews-to-explore) _2026-04-28 · Marco Masi · Quantum Physics: An Overview of a Weird World_ A closer look at Quantum Physics: An Overview of a Weird World ## [The state vector and the Measurement Postulate- Book Excerpt #19](https://quantumworld.substack.com/p/the-state-vector-and-the-measurement) _2026-04-23 · Marco Masi · Quantum Physics: An Overview of a Weird World_ When Reality and Math Meet - Part I ## [SonicMoE: A Hardware-Efficient and Software-Extensible Blueprint for Fine-Grained MoEs](https://tridao.me/blog/2026/sonicmoe-blackwell/) _2026-04-22 · Wentao Guo · Tri Dao_ Figure: SonicMoE's per-layer activation memory footprint (left) stays constant even when expert granularity (embedding dimension / expert intermediate dimension) increases, and SonicMoE can achieve 1.87-4.04x relative speedup compared to existing MoE training kernels ScatterMoE and MoMoE. SonicMoE now runs at peak throughput on NVIDIA Blackwell GPUs (B200/B300), in addition to its existing Hopper… ## [The "Collapse" of the Wave Function - Book Excerpt #18](https://quantumworld.substack.com/p/the-collapse-of-the-wave-function) _2026-04-15 · Marco Masi · Quantum Physics: An Overview of a Weird World_ What is Real? ## [Build with the cheapest GPUs (Sponsored)](https://crawlproof.com/a/198aGEC3yXq3) _2026-04-15 · **Sponsored**_ Deployed a live API in minutes; one user spent under $10. ## [The Wave Function - Book Excerpt #17](https://quantumworld.substack.com/p/the-wave-function-book-excerpt-17) _2026-04-14 · Marco Masi · Quantum Physics: An Overview of a Weird World_ So Abstract and so Real ## [Heisenberg’s Uncertainty Principle - Part III - Book Excerpt #16](https://quantumworld.substack.com/p/heisenbergs-uncertainty-principle-8c9) _2026-04-11 · Marco Masi · Quantum Physics: An Overview of a Weird World_ The Beauty of Uncertainty ## [Gram Newton-Schulz: A Fast, Hardware-Aware Newton-Schulz Algorithm for Muon](https://tridao.me/blog/2026/gram-newton-schulz/) _2026-03-30 · Jack Zhang · Tri Dao_ Muon is becoming the optimizer of choice for training state-of-the-art language models like Kimi K2 Thinking 1 and GLM-5. 2 Compared to AdamW, Muon needs fewer optimizer steps to reach a given loss, but each step is more expensive. This overhead is due to Muon’s Newton-Schulz orthogonalization procedure, a cubic time matrix operation not present in older optimizers. Figure 1: AdamW vs. Muon: Wall… ## [Mamba-3 Part 1](https://tridao.me/blog/2026/mamba3-part1/) _2026-03-16 · Aakash Lahoti\* · Tri Dao_ This series is cross-posted at GoombaLab ## [Mamba-3 Part 2 - Methodological Deep Dive](https://tridao.me/blog/2026/mamba3-part2/) _2026-03-16 · Aakash Lahoti\* · Tri Dao_ This series is cross-posted at GoombaLab ## [FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scaling](https://tridao.me/blog/2026/flash4/) _2026-03-05 · Ted Zadouri · Tri Dao_ \[Paper\] \[Code\]