Links for 2025-07-31

AI

1. GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning https://arxiv.org/abs/2507.19457

2. Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains https://arxiv.org/abs/2507.17746

3. VL-Cogito: Progressive Curriculum Reinforcement Learning for Advanced Multimodal Reasoning https://arxiv.org/abs/2507.22607

4. The Platonic Representation Hypothesis https://arxiv.org/abs/2405.07987

5. Neural networks leverage nominally quantum and post-quantum representations https://arxiv.org/abs/2507.07432

6. New algorithms enable efficient machine learning with symmetric data https://news.mit.edu/2025/new-algorithms-enable-efficient-machine-learning-with-symmetric-data-0730

7. Big study of 100k software devs finds net 15-20% productivity gain from AI. https://www.youtube.com/watch?v=tbDDYKRFjhk

8. Evaluating Grok 4’s Math Capabilities https://epoch.ai/blog/grok-4-math

9. What if AI made the world’s economic growth explode? https://www.economist.com/briefing/2025/07/24/what-if-ai-made-the-worlds-economic-growth-explode [no paywall: https://archive.is/K4F2q]

10. In 2025, "AI capex" - information processing equipment plus software - has added more to US growth than consumer spending. This in spite of the fact that the former is 6% of the economy, and the latter 70%. https://sherwood.news/markets/the-ai-spending-boom-is-eating-the-us-economy/

11. In a new interview, Dario Amodei says that Anthropic has not yet observed any significant diminishing returns when scaling and that he firmly believes that the exponential increase in AI capabilities will continue. https://youtu.be/mYDSSRS-B5U?si=nMdQCa2MPUFkH0bk

12. Tencent releases a model that enables you to generate immersive, explorable, and interactive 3D worlds from just a sentence or an image. Many of the big labs are working in similar directions, including Google. Infinite worlds set to transform game development, VR, digital content creation and so on. https://3d-models.hunyuan.tencent.com/world/

13. China’s Unitree Offers a Humanoid Robot for Under $6,000 https://youtu.be/v1Q4Su54iho

14. Enough AI copilots! We need AI HUDs https://www.geoffreylitt.com/2025/07/27/enough-ai-copilots-we-need-ai-huds

15. Multi-agent systems are the future. https://x.com/pli_cachete/status/1948503774487859466 (See also: Park et al. “Project Sid: Many‑agent simulations toward AI civilization.”, Nov 2024. https://arxiv.org/abs/2411.00114)

16. AlphaEarth Foundations helps map our planet in unprecedented detail https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/

17. Black Forest Labs has released a state-of-the-art open-weights model for text-to-image generation. They say it overcomes the oversaturated 'AI look' with new aesthetics. https://bfl.ai/announcements/flux-1-krea-dev

18. About 30% of Humanity’s Last Exam chemistry/biology answers are likely wrong https://www.lesswrong.com/posts/JANqfGrMyBgcKtGgK/about-30-of-humanity-s-last-exam-chemistry-biology-answers

AI safety

1. Optimizing The Final Output Can Obfuscate CoT (Research Note) https://www.lesswrong.com/posts/CM7AsQoBxDW4vhkP3/optimizing-the-final-output-can-obfuscate-cot-research-note

2. “There exists a parallel track of AI research which has been largely ignored by the AI safety community. This agenda aims to implement human-like online learning in ML models, and it is now close to maturity. Keywords: Hierarchical Reasoning Model, Energy-based Model, Test time training.” https://www.lesswrong.com/posts/tEZa7PouYatK78bbb/i-am-worried-about-near-term-non-llm-ai-developments

3. Safe Artificial General Intelligence through Neuroscience https://blog.amaranth.foundation/p/rfi-neuroscience-and-the-path-to

4. Heeding the Risks of Geopolitical Instability in a Race to Artificial General Intelligence — “In the race to AGI, preventive action could include imposing crippling export controls, poisoning AI training data, or blowing up data centers. The most extreme form of preventive action is a preventive war.” https://www.rand.org/pubs/perspectives/PEA3691-12.html

OpenAI

1. OpenAI researcher Noam Brown on hallucination with the new IMO reasoning model: Mathematicians used to comb through model solutions because earlier systems would quietly flip an inequality or tuck in a wrong step, creating hallucinated answers. Brown says the updated IMO reasoning model now tends to say “I’m not sure” whenever it lacks a valid proof, which sharply cuts down on those hidden errors. https://youtu.be/EEIPtofVe2Q?si=J7VtYt-tGCdR6ClO&t=563

2. “The companies have discussed new terms that would let Microsoft use OpenAI’s latest models and other technology even if the startup decides it has reached its goal of building a more powerful form of AI known as artificial general intelligence (AGI)…” https://www.bloomberg.com/news/articles/2025-07-29/microsoft-s-access-to-openai-tech-is-focus-of-contract-talks [no paywall: https://archive.is/mLEmC]

3. “Stargate Norway is planned to deliver 230MW of capacity, with ambitions to expand by an additional 290MW. The facility will target to deliver 100,000 NVIDIA GPUs by the end of 2026, with the intention to expand significantly in the years ahead.” https://openai.com/index/introducing-stargate-norway/

4. ChatGPT Study Mode https://openai.com/index/chatgpt-study-mode/

Meta

1. Personal Superintelligence https://www.meta.com/superintelligence/

2. Someone at Mira Murati's Thinking Machines turned down a $1 billion offer from Mark Zuckerberg. https://www.wired.com/story/mark-zuckerberg-ai-recruiting-spree-thinking-machines/

---

Biotech

1. Proper embryo selection just landed https://www.emilkirkegaard.com/p/proper-embryo-selection-just-landed

2. “Scientists have successfully grown human livers inside mice—and it could revolutionize how we treat aging. In this episode of Core Memory, we explore the pioneering work at New Limit, a biotech firm pushing the boundaries of epigenetic reprogramming, a groundbreaking technology aiming to rejuvenate old human cells, making them functionally younger and healthier.” https://www.youtube.com/watch?v=GSOSuROez04

3. Scientists genetically edited mosquitoes so they can’t transmit malaria and made the mutation self-replicating. https://www.nature.com/articles/s41586-025-09283-6

Physics and Astronomy

1. Famous double-slit experiment holds up when stripped to its quantum essentials https://news.mit.edu/2025/famous-double-slit-experiment-holds-when-stripped-to-quantum-essentials-0728

2. Early universe’s ‘little red dots’ may be black hole stars https://www.science.org/content/article/early-universe-s-little-red-dots-may-be-black-hole-stars [no paywall: https://archive.is/qRtbe]

3. A deep Search for Ethylene Glycol and Glycolonitrile in the V883 Ori Protoplanetary Disk https://arxiv.org/abs/2507.14905

4. Humans pumping 2 teratons of groundwater shifted Earth's axis of rotation by 80 centimeters https://news.agu.org/press-release/weve-pumped-so-much-groundwater-that-weve-nudged-the-earths-spin/

Neuroscience

1. Benn Jordan converted a drawing of a bird into a spectrogram (PNG -> Soundwave) then played it to a Starling who sung it back, reproducing the PNG. Using the bird’s brain as a hard drive with 2mbps read write speed. https://www.youtube.com/watch?v=hCQCP-5g5bo&t=1019s

2. Triglycerides are an important fuel reserve for synapse function in the brain https://www.nature.com/articles/s42255-025-01321-x

3. “…using deep neural network models that accurately predict hours of brain recordings, we computationally characterise how cortex processes dynamic vision.” https://www.biorxiv.org/content/10.1101/2025.07.22.664908v1

Miscellaneous

1. “Probably this is why God doesn’t connect people’s heart-of-hearts directly to their motor cortex. Instead, He wisely intermediates other brain regions with names like “anterior cingulate gyrus” and “dorsolateral prefrontal area”, the places where rationality happens.” https://www.astralcodexten.com/p/my-heart-of-hearts

2. Facts don't change minds because belief systems work like structural networks https://vasily.cc/blog/facts-dont-change-minds/

3. "On a per capita basis, the highly intelligent became ten times more numerous in England between 1000 and 1850." https://www.aporiamagazine.com/p/the-great-cognitive-advance

4. “Without economic growth, democracy doesn’t work because voters occupy a zero-sum system.” https://blog.samaltman.com/growth-and-government