Links for 2025-07-24
New Temples for the Sandgod
OpenAI and Oracle entered an agreement to develop 4.5 gigawatts of additional Stargate data center capacity, bringing total capacity under development to over 5 gigawatts running over 2 million chips, significantly advancing progress toward their $500 billion commitment announced at the White House in January to invest in 10 gigawatts of AI infrastructure over the next four years.
They are also planning to significantly expand the ambitions of Stargate past the $500 billion commitment.
Read more: https://openai.com/index/stargate-advances-with-partnership-with-oracle/
See also:
After Altman announced 2 million chips, Musk is now talking about 50 million chips. https://x.com/elonmusk/status/1947704195844608094
Anthropic argues that, in order to stay at the frontier of AI development, the United States must be prepared to operate at least 50GW of power capacity for AI workloads by 2028. https://www.anthropic.com/news/build-ai-in-america
Subliminal learning
LLMs transmit traits to other models via hidden signals in data.
Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies.
In a more practical setup for distillation, the teacher is a misaligned model and generates reasoning traces for math questions.
The authors filter out traces that are incorrect or show misalignment.
Yet the student model still becomes misaligned.
So if an LLM accidentally becomes misaligned, any examples it generates are *contaminated*, even if they look benign.
Read more: https://alignment.anthropic.com/2025/subliminal-learning/
Discussion: https://www.lesswrong.com/posts/cGcwQDKAKbQ68BGuR/subliminal-learning-llms-transmit-behavioral-traits-via
Timelines are contracting
When will an AI win a Gold Medal in the International Math Olympiad? Median predicted date over time
July 2021: 2043 (22 years away)
July 2022: 2029 (7 years away)
July 2023: 2028 (5 years away)
July 2024: 2026 (2 years away)
Trump’s ‘AI Action Plan’
Frames AI like the space program of the 1960s: “Just like we won the space race, it is imperative that the United States and its allies win this race.”
It states, “Whoever has the largest AI ecosystem will set global AI standards and reap broad economic and military benefits.”
Announces reforms to green‑light high‑voltage lines, next‑gen nuclear, and geothermal wells so the grid can feed clusters that swallow gigawatts around the clock.
It wants to slash “onerous regulation,” favor open‑source/open‑weight models, speed adoption across industry, government, and DoD, and invest in scientific datasets, automated labs.
The plan acknowledges the risks associated with AI and proposes actions to mitigate them, such as advancing AI interpretability, control systems, and adversarial robustness, as well as evaluating frontier AI systems for security risks and vulnerabilities.
Read more: https://www.whitehouse.gov/articles/2025/07/white-house-unveils-americas-ai-action-plan/
AI
Deep Research Agents with Test-Time Diffusion https://arxiv.org/abs/2507.16075
ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning https://arxiv.org/abs/2507.16815
ASTRO: Teaching Language Models to Reason by Reflecting and Backtracking In-Context https://arxiv.org/abs/2507.00417
Learning without training: The implicit dynamics of in-context learning https://arxiv.org/abs/2507.16003
Apple research just revealed a way to make LLMs 5.35x faster. https://arxiv.org/abs/2507.11851
Inverse Scaling in Test-Time Compute: This study shows that longer reasoning in Large Reasoning Models (LRMs) can hurt performance—revealing a surprising inverse scaling between reasoning length and accuracy. https://safety-research.github.io/inverse-scaling-ttc/
Google's new AI model Aeneas transforms how historians connect the past. Historians found it to be helpful research starting points 9 out of 10 times - improving their confidence by 44%. https://deepmind.google/discover/blog/aeneas-transforms-how-historians-connect-the-past/
LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra https://arxiv.org/abs/2507.15815
Multi-Actor Generative Artificial Intelligence as a Game Engine https://arxiv.org/abs/2507.08892
“Behaviorist” RL reward functions lead to scheming https://www.lesswrong.com/posts/FNJF3SoNiwceAQ69W/behaviorist-rl-reward-functions-lead-to-scheming
Demis Hassabis: Future of AI, Simulating Reality, Physics and Video Games https://lexfridman.com/demis-hassabis-2-transcript
Sam Altman on GPT-5: "This morning I was testing our new model and I got a question. I got emailed a question that I didn't quite understand. And I put it in the model, this is GPT-5, and it answered it perfectly. And I really kind of sat back in my chair, and I was just like, oh man, here it is moment..." https://www.youtube.com/watch?v=aYn8VKW6vXA
ChatGPT users send 2.5 billion prompts a day https://techcrunch.com/2025/07/21/chatgpt-users-send-2-5-billion-prompts-a-day/
34% of U.S. adults have used ChatGPT, about double the share in 2023 https://www.pewresearch.org/short-reads/2025/06/25/34-of-us-adults-have-used-chatgpt-about-double-the-share-in-2023/
Microsoft saved $500 million in its call centers last year by turning to artificial intelligence tools, an executive said. https://www.bloomberg.com/news/articles/2025-07-09/microsoft-using-more-ai-internally-amid-mass-layoffs [no paywall: https://archive.is/uL0um]
Robinhood CEO says the majority of the company's new code is written by AI, with 'close to 100%' adoption from engineers https://finance.yahoo.com/news/robinhood-ceo-says-majority-companys-094801794.html
Abel founder claims Meta offered $1.25 billion over four years to AI hire — 'person still said no' despite equivalent of $312 million yearly salary https://www.tomshardware.com/tech-industry/artificial-intelligence/abel-founder-claims-meta-offered-usd1-25-billion-over-four-years-to-ai-hire-person-still-said-no-despite-equivalent-of-usd312-million-yearly-salary
Detailed list of all 44 people in Meta's Superintelligence team https://www.reddit.com/r/singularity/comments/1m3zn2l/detailed_list_of_all_44_people_in_metas/
Goldman Sachs AI report https://www.goldmansachs.com/what-we-do/investment-banking/insights/articles/powering-the-ai-era/thank-you
Pre-training Auto-regressive Robotic Models with 4D Representations
Robots usually need tons of labeled data to learn precise actions.
What if they could learn control skills directly from human videos… no labels needed?
Robotics pretraining just took a BIG jump forward.
A new Autoregressive Robotic Model, learns low-level 4D representations from human video data.
Bridging the gap between vision and real-world robotic control.
Why this matters:
✅ Pretraining with 4D geometry enables better transfer from human video to robot actions
✅ Overcomes the gap between high-level VLA pretraining and low-level robotic control
✅ Unlocks more accurate, data-efficient learning for real-world tasks
Paper and Code: https://arm4r.github.io/
A real-world study of an LLM clinical copilot
Across 39,849 live patient visits, clinicians with AI had a 16% relative reduction in diagnostic errors and a 13% reduction in treatment errors vs. those without. 100% of survey respondents said it improved their quality of care, and clinicians reported it broadened their knowledge.
Paper: https://openai.com/index/ai-clinical-copilot-penda-health/
Related studies:
1. The largest medical AI randomized controlled trial yet performed, enrolling >100,000 women undergoing mammography screening, was just published.
The use of AI led to 29% higher detection of cancer, no increase of false positives, and reduced workload compared with radiologists without AI.
Paper: https://thelancet.com/journals/landig/article/PIIS2589-7500(24)00267-X/fulltext
2. Superhuman AI at medical diagnosis: It’s so good that clinicians do *worse* when assisted by the AI compared to if they just let the AI do its job.
Paper 1: https://goo.gle/4lpQ8xg
Paper 2: https://goo.gle/3G4DNPe
In the first paper, AMIE matched or exceeded primary care physicians in a study of simulated text-chat medical consultations with validated patient actors.
The second paper evaluated AMIE’s ability to generate a differential diagnosis alone or as a clinician aide. The authors looked at 303 challenging, real-world medical cases. AMIE outperformed clinicians, and clinicians assisted by AMIE showed improved performance.
A generic non-invasive neuromotor interface for human-computer interaction
Advanced ML models and EMG hardware can transform neural signals controlling muscles at the wrist into commands that seamlessly drive computer interactions.
Paper: https://www.nature.com/articles/s41586-025-09255-w
Tracking Advances in Neural Recording
Over the last five decades, progress in neural recording has enabled the number of simultaneously recorded neurons to approximately double every 7 years, mirroring Moore’s Law.
Read more: https://stevenson.lab.uconn.edu/scaling/
Science and Technology
Studies show that the obesity and diabetes medication also reduces heart attacks, cancer risk, migraines, and memory loss. How is that even possible? And at what point should we all be on it? https://www.derekthompson.org/p/why-does-it-seem-like-glp-1-drugs
"An adult mouse brain, conservatively estimated to be 1 cm^3, imaged at 15 nm isotropic resolution, as tentatively demonstrated above, would result in a dataset of 2.96 × 10^17 voxels. A single PEEM with a CW UV laser, imaging non-stop at 2 Gigahertz pixel rate, would take approximately five years to finish that acquisition." https://biorxiv.org/content/10.1101/2023.09.05.556423v3.full
Natural sounds can be reconstructed from human neuroimaging data using deep neural network representation https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3003293
Advances in DNA-synthesis technologies are allowing large volumes of DNA to be pumped out a low cost https://www.nature.com/articles/d41586-025-02261-y [no paywall: https://archive.is/koBG2]
When Betelgeuse Explodes, It’s Going to Take Out Another Star https://www.nytimes.com/2025/07/21/science/betelgeuse-star-companion.html [no paywall: https://archive.is/mO2N6]
Scientists overwhelmingly recognize the value of sharing null results, but rarely publish them in the research literature https://www.nature.com/articles/d41586-025-02312-4 [no paywall: https://archive.is/AjWXQ]
Math
Equations Mean Things https://www.lesswrong.com/posts/FR2aPsEcaRfy7iKZW/equations-mean-things
Consonance, dissonance, eigenvalue spectra and other universes https://aleph.se/andart2/uncategorized/consonance-dissonance-eigenvalue-spectra-and-other-universes/
Ukraine
The nationwide protests against Zelenskyy over his actions against independent anti-corruption bodies again show what distinguishes Ukraine from Russia. Ukraine has a civil society that is ready to stand up for its values, even during wartime. As with the Orange Revolution in 2004 and Euromaidan in 2014, the main concern of Ukrainians is transforming Ukraine into a modern nation prepared to join the European Union.
It's hard to exaggerate how unpopular Zelenskyy's recent actions have been. Every unofficial Ukrainian military channel, soldier, and activist I'm aware of has condemned his actions and welcomed criticism from Western countries.
A short summary of what's going on:
Ukraine created independent anti-corruption institutions (NABU, SAPO, etc.) whose leaders are chosen by mixed commissions with Ukrainian members and independent international experts (an EU-backed safeguard for autonomy).
Those bodies worked outside the traditional hierarchy, allowing them to go after the system itself. For example, when a corrupt official is investigated, he can’t just call his guy at the top to shut it down. The agency doesn’t answer to the president or a minister. That’s the point. They have been probing high-level officials, angering parts of the government.
These independent institutions are not a detail of reform. They are the foundation of democratic Ukraine.
The Zelenskyy Cabinet blocked one of the candidates, citing “security” concerns.
Authorities raided NABU/SAPO and charged activist Vitaliy Shabunin, prompting accusations of political pressure.
Parliament passed, and Zelenskyy signed, a law subordinating NABU/SAPO to the Prosecutor General, triggering protests and EU/G7 criticism.
Zelenskyy now promises a new bill, but trust has been damaged.
Related:
Putin launches spy app to keep Russians in ‘digital gulag’ https://www.thetimes.com/world/russia-ukraine-war/article/putin-moscow-whatsapp-ban-plan-max-app-launch-b789tt6ts [no paywall: https://archive.is/zncUF]
“Just few days ago, I mapped 450 new russian airstrikes in 8 days, north of Pokrovsk.” https://x.com/clement_molin/status/1948112638061252748