Links for 2026-01-28
AI
Helix 02: Full-Body Autonomy https://www.figure.ai/news/helix-02
“The Microsoft Superintelligence team will use Maia 200 for synthetic data generation and reinforcement learning to improve next-generation in-house models. For synthetic data pipeline use cases, Maia 200’s unique design helps accelerate the rate at which high-quality, domain-specific data can be generated and filtered, feeding downstream training with fresher, more targeted signals.” https://blogs.microsoft.com/blog/2026/01/26/maia-200-the-ai-accelerator-built-for-inference/
Researchers at the Flatiron Institute have developed a new biologically inspired computational unit for artificial neural networks called “rectified spectral units” (ReSUs), which extract temporal patterns from the recent past to predict the near future. Unlike standard units that rely on biologically implausible backpropagation, these self-supervised ReSUs successfully replicated complex features of the fruit fly visual system by learning to process raw data independently. https://www.simonsfoundation.org/2026/01/26/biological-brains-inspire-a-new-building-block-for-artificial-neural-networks/
New essay by Anthropic CEO Dario Amodei: We are considerably closer to real danger in 2026 than we were in 2023. AIs will have a fairly good shot at taking over the world and imposing its will on everyone else. The idea of stopping or substantially slowing AI is fundamentally untenable. https://www.lesswrong.com/posts/kzPQohJakutbtFPcf/dario-amodei-the-adolescence-of-technology
Andrej Karpathy: “I really am mostly programming in English now” https://x.com/karpathy/status/2015883857489522876
Numina-Lean-Agent: An Open and General Agentic Reasoning System for Formal Mathematics https://arxiv.org/abs/2601.14027
AI can now find real security vulnerabilities in the most hardened, well-audited codebases on the planet. https://www.lesswrong.com/posts/7aJwgbMEiKq5egQbd/ai-found-12-of-12-openssl-zero-days-while-curl-cancelled-its
A concrete demo of AI-assisted formal verification becoming practical: Formalizing serious crypto proofs is getting cheaper/faster. https://blog.zksecurity.xyz/posts/simple-rbr-fri/
Problem 728 and the use of AI on Erdős problems https://www.erdosproblems.com/forum/thread/blog:2
EpochAI released a new benchmark: A collection of unsolved mathematics problems that have resisted serious attempts by professional mathematicians. https://epoch.ai/frontiermath/open-problems
Prism: a free workspace for scientists to write and collaborate on research, powered by GPT-5.2. http://prism.openai.com/
AlgZoo: A collection of extremely tiny AI models (ranging from 8 to roughly 1,400 parameters) that serve as a “stress test” for our ability to understand how AI actually works. https://www.lesswrong.com/posts/x8BbjZqooS4LFXS8Z/algzoo-uninterpreted-models-with-fewer-than-1-500-parameters
When open-source models are fine-tuned on seemingly benign chemical synthesis information generated by frontier models, they become much better at chemical weapons tasks. https://www.arxiv.org/abs/2601.13528
Self-Distilled Reasoner: On-Policy Self-Distillation https://siyan-zhao.github.io/blog/2026/opsd/
A Pragmatic VLA Foundation Model https://arxiv.org/abs/2601.18692
A case study of a “model breakdown,” where a routine request triggered a deep recursive loop in the model’s internal logic, resulting in multilingual gibberish and a fixation on seemingly random concepts like tumors and Jose Mourinho. https://www.lesswrong.com/posts/XuzPu5mBDY3TCvw2J/anomalous-tokens-on-gemini-3-0-pro
China Trains AI-Controlled Weapons With Learning From Hawks, Coyotes https://www.wsj.com/world/china/china-ai-weapons-hawks-wolves-2fcb58bb [no paywall: https://archive.is/8EyYr]
Xi Jinping calls AI ‘epoch-making’ as China pushes innovation strategy – but flags risks https://www.scmp.com/economy/china-economy/article/3341267/xi-jinping-calls-ai-epoch-making-china-pushes-innovation-strategy-flags-risks
Beijing approves the purchase of “several hundred thousand” Nvidia H200 chips worth around $10 billion for large tech companies like Alibaba and ByteDance. More approvals are expected later. https://www.reuters.com/world/china/china-gives-green-light-importing-first-batch-nvidias-h200-ai-chips-sources-say-2026-01-28/ [no paywall: https://archive.is/RVQOh]
NATO plans to create an automated robotic defense zone bordering Belarus and Russia manned with a network of armed drones, turrets, quadruped robots, semi-autonomous combat vehicles, unmanned robotic ground systems, and automated air and missile defense systems. [German language article] https://www.welt.de/politik/ausland/article696de16a18af1f04e38dd6da/nato-general-ziel-ist-es-den-feindlichen-vorstoss-zu-brechen-nato-will-an-der-ostflanke-massiv-aufruesten.html [no paywall: https://archive.is/XcFJ6]
Hyundai Motor’s Korean union warns of humanoid robot plan, sees threat to jobs https://www.reuters.com/business/world-at-work/hyundai-motors-korean-union-warns-humanoid-robot-plan-sees-threat-jobs-2026-01-22/ [no paywall: https://archive.is/G1V4E]
A January 2026 report from Georgetown’s CSET confirms that Google DeepMind and OpenAI are pursuing “recursive self-improvement” (RSI), a paradigm that could exponentially accelerate AI capabilities beyond human control. https://cset.georgetown.edu/publication/when-ai-builds-ai/
Economist: AI will be bigger than electricity and semiconductors [PDF] https://web.stanford.edu/~chadj/AIandEconomicFuture.pdf
AI Now Beats the Average Human in Tests of Creativity https://singularityhub.com/2026/01/27/ai-now-beats-the-average-human-in-tests-of-creativity/
Dialogue: Is there a Natural Abstraction of Good? https://www.lesswrong.com/posts/M5s6WgScRfmeWsLD4/dialogue-is-there-a-natural-abstraction-of-good
Brain emulation
Notable Progress Has Been Made in Whole Brain Emulation https://www.lesswrong.com/posts/DGsBfcEQKuNPmQizQ/notable-progress-has-been-made-in-whole-brain-emulation
State of Brain Emulation 2025 https://brainemulation.mxschons.com/
Building Brains on a Computer https://press.asimov.com/articles/brains