632nm — searchable YouTube transcripts & AI Q&A

Google's AlphaQubit: Can RL Fix Quantum Errors? | Vlad Sivak Google Quantum AI

Vlad Sivak argues that reinforcement learning can improve quantum error correction by calibrating hardware and decoding syndromes, as demonstrated by their work at Google Quantum AI. He explains that RL is model-free and can adapt to system drift, unlike traditional calibration methods, and that it can be scaled using the locality of error syndromes. However, he notes that RL is not yet capable of calibrating a system from scratch and relies on a good starting point.

Where Does Intelligence Live? | Michael Levin on Evolution, AI, and Platonic Space

Michael Levin argues that intelligence and cognitive competencies are not solely products of evolution, engineering, or learning, but are patterns that exist in a Platonic space of mathematical truths and can ingress into physical systems through interfaces. He illustrates this with examples like planarians adapting to barium by changing fewer than a dozen genes, and random gene regulatory networks exhibiting learning without selection. He suggests that understanding these free lunches could transform medicine and AI, but emphasizes the need for ethical consideration of the minds that may emerge.

Inside the AI-Driven mRNA Cancer Vaccine Revolution | Sid Sijbrandij on GitLab + Biotech

Sid Sijbrandij argues that cancer patients should consider combining multiple experimental treatments rather than relying solely on standard of care, which is often limited by liability concerns and trial designs that prioritize data collection over patient benefit. He shares his own experience of using five immunotherapies in parallel, a radiopharmaceutical, and an mRNA vaccine, which led to a 15-month cancer-free period, though he cautions that this approach is risky and not suitable for everyone. He also emphasizes the importance of open-sourcing medical data and starting companies to make personalized treatments more accessible and affordable.

Building Quantum Computers at Semiconductor Scale | John Martinis on Qolab + Superconducting Qubits

John Martinis argues that the key to scaling superconducting quantum computers is moving away from academic fabrication techniques like liftoff and instead using modern semiconductor manufacturing processes, such as deposition and etch, to achieve the reliability and yield needed for millions of qubits. He explains that while current devices with around 100-200 qubits work, scaling up requires solving system-level engineering challenges like wiring, power dissipation, and the presence of two-level state defects, which his company Qolab aims to address. He also reflects on the Google quantum supremacy experiment, emphasizing that it was a fundamental test of quantum mechanics and that the field's next goal is to achieve useful quantum computation.

Diffraction Limit, Microscopy, and Cell Biology | Eric Betzig on Super-Resolution Microscopy

Eric Betzig explains the diffraction limit in microscopy, its impact on cell biology, and his journey to super-resolution microscopy.