Amazon Science’s Post

Very excited to share our work on "Science in the Age of Foundation Models", where we discuss applications of foundation models in time series forecasting and PDEs including weather forecasting and aerodynamics.

Google's Willow processor just achieved what physicists pursued for 30 years—exponential quantum error reduction. Microsoft integrated control electronics directly onto quantum chips. D-Wave demonstrated computational supremacy in real-world physics. But here's what $10B valuations and 5,000+ qubit systems obscure: quantum computing's bottleneck isn't hardware anymore. It's data quality. New research reveals stunning dependencies on annotation precision: IonQ-Roche found 3% annotation improvement = 42% better drug candidate identification JPMorganChase discovered 3% better labels = 0.23 Sharpe ratio increase (worth billions) Toyota Motor Corporation achieved 28% performance gains with temperature-stratified annotation protocols The most surprising finding? Grad students using rewards-based and QC'd app platforms outperform senior professors by 12% in accuracy and 350% in throughput. The top 5% of annotators contribute 31% of total value—a massive power law distribution. Why does this matter to you? For VCs: The quantum market reaches $7.3B by 2030, but winners will be determined by data quality infrastructure, not just qubit counts. Companies using commodity annotation need 5x more training data and 6 additional months per algorithm. For Scientists: Your quantum ML models are only as good as your training data. A single mislabeled quantum state can cascade into thousands of logical operation failures. Expert disagreement (15-20%) actually IMPROVES model robustness by 23%. For Corporations: McKinsey & Company projects $850B value unlock by 2035, but only for those who invest in expert annotation now. It accelerates quantum advantage by 2-3 years. First movers in pharmaceuticals already save $1.1B per approved drug. The revolution won't be limited by qubit counts or coherence times—it will be enabled or constrained by annotation quality. As IonQ's Chris Monroe stated: "We've solved the physics. Now we need to solve the data science." Read the full analysis including case studies from Google, Microsoft, IonQ, Quantinuum, and 40+ quantum leaders, plus detailed ROI calculations: https://lnkd.in/gkV6HQ_U What's your take—will Data Quality, ML Models or Hardware determine quantum computing winners? #QuantumComputing #DeepTech #DataScience #QuantumML #DataLabeling #QuantumAI #QuantumDataLabeling #VentureCapital

I explored how quantum computing might rewrite the rules of machine learning. QRF was a speculative project—reimagining decision trees through quantum logic like superposition and entanglement. I documented it thoroughly, wrote prototype code, and translated cutting-edge research into a readable, testable form. Business value: It's not about deploying today—it’s about being ready for what’s next. Why it matters: I showed I can self-learn advanced domains, synthesize them into code, and lead the conversation before the market catches up.

Today, one of Microsoft Research's teams in Cambridge, UK published a breakthrough result in Nature Magazine. They disclosed a new type of computer that can solve hard, complex optimization problems almost entirely in the analog optical domain. This announcement is the result of many years of hard work by the team, combining sophisticated mathematical theory with innovations in hardware and optics. This new system addresses previously intractable optimization challenges, with numerous real-world examples disclosed, including reducing the time to do MRI scans by 6x. But beyond individual applications, the technology may allow major jumps in AI by finding much better optimization points in a computationally tractable manner. Congratulations to Hitesh Ballani, Francesca Parmigiani, and the rest of the team for this announcement. It's an excellent example of why Microsoft Research trusts its people to choose their research directions, and encourages them to explore new spaces whose value may not be immediately apparent. If you are interested, read more here: https://lnkd.in/gT_9Sgnu

🚨 Why Mira Murati’s Breakthrough Matters — and What It Reveals About LLMs Today’s LLMs aren’t just sometimes wrong — they’re also not reproducible. Ask the same question 5 times and get 5 different answers — even with “deterministic” settings. Murati’s new lab, Thinking Machines, just showed why: GPU kernels used in inference aren’t batch-invariant. 🧠 In plain terms: if multiple users send prompts at once, the server batches them for efficiency. Tiny differences in operation order cause floating-point shifts — which cascade into different neurons and answers. If your request runs alone, you might avoid this. But in production, they almost never do. Their fix? New kernels (matmul, attention, RMSNorm) that guarantee: same input → same output, every time. Why this matters: • Compliance: Reproducibility is mandatory in regulated sectors. • Cost: Stable outputs enable caching, cutting GPU burn. •Productization: Expect enterprise “deterministic modes.” ⚠️ But: determinism ≠ correctness. You may now get the same wrong answer every time. Reliability needs both consistency and correctness. https://lnkd.in/gYPaE2S6

Some really interesting research from the team in Cambridge! In a breakthrough, a Microsoft analog optical computer–a machine that uses light as a computing medium–has solved two complex optimization problems in healthcare and finance and has shown it has potential for AI workloads using much less energy. The work of our research team appears in a paper in the journal Nature. https://msft.it/6041s5y0P

When Computers Start Thinking in Light - Microsoft’s researchers in Cambridge have unveiled a prototype optical computer that trades electrons for beams of light—and in doing so, may have leapt past silicon’s limits. This analog optical machine doesn’t crunch binary code; it lets the physics of light itself perform the math, enabling calculations up to 100 times faster with a fraction of the energy. Already tested on complex financial simulations, the system hints at a future where AI, medicine, and markets run on light-speed logic, rewriting the very foundation of computation. https://lnkd.in/egFgnW5w FuturistSpeaker.com

KEYNOTE 2: Generative AI in Healthcare – a Socio-Technical Challenge for Computer Scientists Designing systems that are not only powerful, but also safe, explainable, and clinically meaningful Professor Ping Yu – School of Computing and Information Technology, University of Wollongong, Australia #HIKM2025

LLMs don’t give the same answer to the same input. Until now. Like a calculator that sometimes says 2+2 is five, depending on when you hit the keys, LLMs have long behaved in ways we came to accept as unpredictable. Now, Mira Murati’s team at Thinking Machines Lab has fixed it. The issue was not “randomness” or “creativity,” but noise in the infrastructure. Tiny nondeterministic effects inside GPU kernels ripple forward and change completions even when the input is the same. Their fix, naturally explained in jargon, was to design batch-invariant kernels for matmul, attention, and RMSNorm, and the bottom line is clear: same input equals same output, every time. Why it matters: - Reliability in high-stakes fields: health and finance cannot accept answers that drift with server load. - Operational savings: deterministic outputs mean caching works, cutting GPU burn. - Open source transparency: the fix is published for anyone to use and build on. This does not solve the correctness of answers, since wrong outputs can still be wrong consistently. But it clears away one of the biggest barriers to making LLMs reliable at scale. Just days ago this was another open mystery in how LLMs work. Now it is progress — another win for science and engineering. Read more here: https://lnkd.in/d7iNCjzj