After losing her hands to meningitis as a baby, Tilly has become the first person in the world to receive fully wireless bionic hands, giving her a whole new level of independence. 🦾

Source: https://www.instagram.com/reel/DIg6badoMUI/

Links for 2025-06-27 [Part 1]

AI

1. Steering Your Diffusion Policy with Latent Space Reinforcement Learning — “If you have a policy that uses diffusion/flow (e.g. diffusion VLA), you can run RL where the actor chooses the noise, which is then denoised by the policy to produce an action. This method, which we call diffusion steering (DSRL), leads to a remarkably efficient RL method!” https://diffusion-steering.github.io/

2. “We show how to steer DeepSeek-R1-Distill’s reasoning: make it backtrack, add knowledge, test examples. Just by adding steering vectors to its activations!” https://arxiv.org/abs/2506.18167

3. Language Models use Lookbacks to Track Beliefs https://belief.baulab.info/

4. Just Enough Thinking: Efficient Reasoning with Adaptive Length Penalties Reinforcement Learning https://arxiv.org/abs/2506.05256

5. Bridging Offline and Online Reinforcement Learning for LLMs https://arxiv.org/abs/2506.21495

6. “Sparse attention (MoBA/NSA) trains faster & beats full attention in key tasks. But we’ve had no idea how they truly work…until now.” https://www.tilderesearch.com/blog/sparse-attn

7. AXIOM: Mastering Arcade Games in Minutes with Active Inference and Structure Learning https://www.verses.ai/research-blog/axiom-mastering-arcade-games-in-minutes-with-active-inference-and-structure-learning

8. AI Improves at Improving Itself Using an Evolutionary Trick https://spectrum.ieee.org/evolutionary-ai-coding-agents

9. “We found a way of bridging the gap between the existing powerful implementations and scalability of modern AI, and the biological plausibility paradigm where neuron timing matters.” https://pub.sakana.ai/ctm/

10. When Life Gives You Samples: The Benefits of Scaling up Inference Compute for Multilingual LLMs https://arxiv.org/abs/2506.20544

11. What does 10x-ing effective compute get you? https://www.lesswrong.com/posts/hpjj4JgRw9akLMRu5/what-does-10x-ing-effective-compute-get-you

12. “This post lays out how the industrial explosion could play out, and how quickly it might happen.” https://www.lesswrong.com/posts/Na2CBmNY7otypEmto/the-industrial-explosion

13. Why Aren't LLMs General Intelligence Yet? https://minihf.com/posts/2025-06-25-why-arent-llms-general-intelligence-yet/

14. An Agentic System for Rare Disease Diagnosis with Traceable Reasoning (achieving 100% accuracy for 1013 diseases) https://arxiv.org/abs/2506.20430

15. The Dream of an AI Scientist Is Closer Than Ever https://singularityhub.com/2025/06/26/the-dream-of-an-ai-scientist-is-closer-than-ever/

16. Trump plans executive orders to power AI growth in race with China https://www.reuters.com/legal/government/trump-plans-executive-orders-power-ai-growth-race-with-china-2025-06-27/ [no paywall: https://archive.is/GFAgl]

17. Spanish mathematician Javier Gómez Serrano and Google DeepMind team up to solve the Navier-Stokes million-dollar problem https://english.elpais.com/science-tech/2025-06-24/spanish-mathematician-javier-gomez-serrano-and-google-deepmind-team-up-to-solve-the-navier-stokes-million-dollar-problem.html

18. Satya Nadella: We're One Algorithm Away from the Next AI Breakthrough https://youtu.be/AUUZuzVHKdo?si=Xxs0ZPWH0g6AMJOL&t=942

19. One of the Best Hackers in the Country is an AI Bot https://www.bloomberg.com/news/articles/2025-06-24/one-of-the-best-hackers-in-the-country-is-an-ai-bot [no paywall: https://archive.is/luu65]

20. Only 32 nations, mostly in the Northern Hemisphere, have A.I.-specialized data centers. https://www.nytimes.com/interactive/2025/06/23/technology/ai-computing-global-divide.html [no paywall: https://archive.is/WvrC5]

Links for 2025-06-27 [Part 2]

AI

21. Meta hires key OpenAI researcher to work on AI reasoning models https://techcrunch.com/2025/06/26/meta-hires-key-openai-researcher-to-work-on-ai-reasoning-models/

22. Mark Zuckerberg has successfully lured away three OpenAI researchers. Lucas Beyer, Alexander Kolesnikov and Xiaohua Zhai have joined META's superintelligence lab. https://www.wsj.com/tech/ai/meta-poaches-three-openai-researchers-eb55eea9 [no paywall: https://archive.is/mDNUP]

23. Meta wins AI copyright case, but judge says others could bring lawsuits https://www.cnbc.com/2025/06/25/meta-llama-ai-copyright-ruling.html

24. "Chatbots have proved adept at keeping users locked into conversations. … by answering every question …poses an existential threat to the media, and to the livelihood of journalists everywhere." https://www.theatlantic.com/technology/archive/2025/06/generative-ai-pirated-articles-books/683009/ [no paywall: https://archive.is/rJIBT]

25. “We had Claude run a small shop in our office lunchroom. Here’s how it went.” https://www.anthropic.com/research/project-vend-1

26. Thought Anchors: Which LLM Reasoning Steps Matter? https://arxiv.org/abs/2506.19143

27. Orbformer, a foundation model for wavefunctions using deep QMC that offers a route to tackle strongly correlated quantum states https://ae-foster.github.io/posts/2025/06/25/orbformer-post.html

Brains

1. How mice generalize knowledge across vision and touch, thanks to an abstract representation of peripersonal space in their dorsal cortex. https://www.nature.com/articles/s41467-025-59342-9

2. How do neural dynamics in motor cortex interact with those in subcortical networks to flexibly control movement? Cortical sculpting of a rhythmic motor program https://www.biorxiv.org/content/10.1101/2025.06.20.660772v1

3. Scientists complete largest wiring diagram and functional map of the brain to date https://alleninstitute.org/news/scientists-complete-largest-wiring-diagram-and-functional-map-of-the-brain-to-date/

Bio(tech)

1. Interview with George Church, the godfather of modern synthetic biology. He thinks that the last few decades of progress have put us on the verge of the big payoffs: deaging, deextinction, biobots that combine the best of human & natural engineering, and (unfortunately) weaponized mirror life. https://www.dwarkesh.com/p/george-church

2. Missing Heritability: Much More Than You Wanted To Know https://www.astralcodexten.com/p/missing-heritability-much-more-than

3. Heritability of achievement tests at age 10-14 in Norway's entire population https://www.emilkirkegaard.com/p/heritability-of-achievement-tests

Miscellaneous

1. “Our findings provide strong evidence of a partisan asymmetry in misinformat” https://marginalrevolution.com/marginalrevolution/2025/06/the-objectivity-of-community-notes.html

2. The World Is Warming Up. And It’s Happening Faster. https://www.nytimes.com/2025/06/26/climate/climate-heat-intensity.html [no paywall: https://archive.is/Jn9Dm]

LLM context windows have grown, but can models really use all this content?

Since mid-2023, the longest LLM context windows have grown by about 30x per year. Their ability to use that input effectively is improving even faster: on two long-context benchmarks, the input length where top models reach 80% accuracy has risen by over 250x in the past 9 months.

Read more: https://epoch.ai/data-insights/context-windows

In case you missed it, Mark Zuckerberg has suddenly gone all‑in on superintelligence.

Just within the past couple of weeks:

1. Meta created a dedicated superintelligence lab.
2. It invested $14.3 billion to acquire a 49 % stake in a data‑labeling firm.
3. It launched a hostile talent‑acquisition frenzy, poaching top talent from OpenAI.

Meta aims to deploy over 1.3 million GPUs by the end of the year.

Higher adolescent cognitive ability is a broad, dose-dependent protective factor against most mid-life mental-disorder diagnoses

Even when all shared family factors (genes and home environment) are considered, a link between lower cognitive ability and a higher risk of receiving a mental health diagnosis persists.

The authors did not find any evidence for the idea that having “too much” cognitive ability is bad for your mental health.

Note: In Model 1, every step (≈ 7.5 IQ points) lowers the odds of a diagnosis by ~16 %.

Study is robust:

- Huge sample: 272,351 men
- Nationwide administrative data
- Rather than relying on self-reporting, diagnoses are taken from the national register for primary care physicians (95.3%) and emergency room consultations (4.7%).
- Cognitive scores from mandatory military screening.

Paper: https://journals.sagepub.com/doi/10.1177/09567976251347221

If you want evidence that the big AI labs truly believe that AI investments are worth it, consider how much money they spend on improving capabilities that the vast majority of users don't care about. Almost nobody cares about solving research math, protein folding, or genome prediction.

In other words, user adoption isn't their primary metric. External adoption funds R&D, but the ultimate value comes from internal leverage. As demonstrated by Google, their true goal is to develop AI models, such as AlphaChip and AlphaEvolve, that improve their chips and algorithms. They don't even give other parties access to these models. They are optimized for strategic advantage. Their goal is self-improvement.

Even things like the Veo 3, which average people find exciting, were not designed for them. Video generation is an early ingredient of world models that learn physics and causality, foundations useful for robotics and planning.

OpenAI Leadership Responds to Meta Offers: ‘Someone Has Broken Into Our Home’

Read more: https://www.wired.com/story/openai-meta-leadership-talent-rivalry/ [no paywall: https://archive.is/mdCIi]

Links for 2025-06-30 [Part 1]

AI

1. Chai-2, a major breakthrough in molecular design. Chai-2 enables zero-shot antibody discovery in a 24-well plate, exceeding previous SOTA by >100x. Beyond antibodies, Chai-2 demonstrates a 68% wet-lab hit rate in miniprotein binder design, routinely yielding picomolar affinities. [PDF] https://chaiassets.com/chai-2/paper/technical_report.pdf

2. Researchers Uncover Hidden Ingredients Behind AI Creativity https://www.quantamagazine.org/researchers-uncover-hidden-ingredients-behind-ai-creativity-20250630/

3. Accelerating scientific discovery with AI https://news.mit.edu/2025/futurehouse-accelerates-scientific-discovery-with-ai-0630

4. Tesla achieves autonomous car delivery from factory to customer https://www.theverge.com/news/694801/tesla-autonomous-delivery-factory-customer-robotaxi

5. DemoDiffusion: An extremely simple approach enabling a pre-trained 'generalist' diffusion policy to follow a human-demonstration for a novel task during inference. One-shot human imitation *without* requiring any paired human-robot data or online RL. https://demodiffusion.github.io/

6. WorldVLA: Towards Autoregressive Action World Model https://arxiv.org/abs/2506.21539

7. Discovering Temporal Structure: An Overview of Hierarchical Reinforcement Learning https://www.arxiv.org/abs/2506.14045

8. MiCo: Multi-image Contrast for Reinforcement Visual Reasoning https://arxiv.org/abs/2506.22434

9. Stochastic Parameter Decomposition https://www.lesswrong.com/posts/yjrpmCmqurDmbMztW/paper-stochastic-parameter-decomposition

10. GPAS: Accelerating Convergence of LLM Pretraining via Gradient-Preserving Activation Scaling https://arxiv.org/abs/2506.22049

11. Radial Attention: O(nlogn) Sparse Attention with Energy Decay for Long Video Generation https://arxiv.org/abs/2506.19852

12. Google presents Performance Prediction for Large Systems via Text-to-Text Regression https://arxiv.org/abs/2506.21718

13. The Automated LLM Speedrunning Benchmark: Reproducing NanoGPT Improvements https://arxiv.org/abs/2506.22419

14. Apple has asked both Anthropic and OpenAI to train versions of their models to run on Apple's cloud infrastructure. Both models will be tested, and the winner of this duel will power the new Siri. If Anthropic wins then Claude will power both Alexa and Siri. https://www.bloomberg.com/news/articles/2025-06-30/apple-weighs-replacing-siri-s-ai-llms-with-anthropic-claude-or-openai-chatgpt [no paywall: https://archive.is/0786n]

15. UK entry-level jobs have dropped 32% since ChatGPT launch https://www.theguardian.com/business/2025/jun/30/uk-entry-level-jobs-chatgpt-launch-adzuna

16. What you can do about AI 2027 https://blog.ai-futures.org/p/what-you-can-do-about-ai-2027

17. “In key centers of power, there’s an important shift happening now of the Overton Window for AI dangers.” https://bayesianinvestor.com/blog/index.php/2025/06/29/waking-up-to-agi/

18. An amendment would add notable exceptions to the GOP’s prohibition on state AI laws. https://www.politico.com/live-updates/2025/06/29/congress/blackburn-cruz-find-potential-truce-on-state-ai-moratorium-child-safety-00432296

19. The Folly of "Legal" AI Products https://cashandcarried.substack.com/p/the-folly-of-legal-ai-products

20. “these guys are only 2 weeks old, 411k spotify listeners. & entirely ai generated. this is the first time i’ve seen a totally synthetic act hit cultural velocity this fast.” https://www.youtube.com/watch?v=eQJ9IWoclhk

Links for 2025-06-30 [Part 2]

Energy

1. Major Demo Keeps Quaise Energy on Track to Power the World with Clean, Renewable Geothermal Energy https://www.quaise.energy/news/major-demo-keeps-quaise-energy-on-track-to-power-the-world-with-clean-renewable-geothermal-energy

2. Google is betting on a fusion-powered future https://blog.google/outreach-initiatives/sustainability/our-latest-bet-on-a-fusion-powered-future/

Computer Science

1. Inside the British lab growing a biological computer https://www.ft.com/content/713eab47-a1f1-4477-a7de-f2b150e2faac [no paywall: https://archive.is/DeOn8]

2. New Proof Dramatically Compresses Space Needed for Computation https://www.scientificamerican.com/article/new-proof-dramatically-compresses-space-needed-for-computation/

3. BusyBeaver(6) is really quite large https://scottaaronson.blog/?p=8972

Biotech

1. Customizable virus-like particles deliver CRISPR–Cas9 ribonucleoprotein for effective ocular neovascular and Huntington’s disease gene therapy https://www.nature.com/articles/s41565-024-01851-7

2. Scientists Genetically Engineer Tobacco Plants to Pump Out a Popular Cancer Drug https://singularityhub.com/2025/06/30/scientists-genetically-engineer-tobacco-plants-to-pump-out-a-popular-cancer-drug/

Miscellaneous

1. Mysterious link between Earth’s magnetism and oxygen levels baffles scientists https://www.nature.com/articles/d41586-025-01862-x [archived version: https://archive.is/zYBBD]

2. "A formal proof of the Born rule from decision-theoretic assumptions", Wallace 2009 https://arxiv.org/abs/0906.2718

3. Peter Putnam (1927–1987): forgotten early philosopher of model-free RL / predictive processing neuroscience https://nautil.us/finding-peter-putnam-1218035/

In May 2025, Epoch AI convened 30 of the world's top mathematicians in Berkeley for a two-day retreat, marking the capstone of FrontierMath Tier 4.

The symposium participants and judges tested AI models on their most challenging math problems, and discussed AI’s future in math. This video of the closing ceremony offers a glimpse into the symposium's collaborative atmosphere and dedication of the participants behind crafting the hardest mathematical test for AI ever made.

-- Timestamps --
00:00 - Intro by Elliot Glazer - Epoch AI
03:38 - Topology | Sergei Gukov - Caltech
06:53 - Algebraic Geometry | Ravi Vakil - Stanford
09:10 - Number Theory | Ken Ono - University of Virginia
13:19 - Combinatorics | Igor Pak - UCLA
16:54 - Analysis | Paata Ivanisvili - UC Irvine
18:18 - Closing Remarks

SPIRAL: A fully-online, multi-turn, multi-agent RL system that can finetune the entire LLM during live self-play.

A key innovation of SPIRAL is the use of self-play in games to generate an endless supply of progressively harder problems, as the model must constantly adapt to a more capable version of itself.

Highlights:

* An automatic curriculum: as the model improves, so does its opponent, so the problems keep getting harder without any human data or rewards.

* Reasoning gains from a single game: Training Qwen-3 4B only on Kuhn Poker boosts math benchmarks by +8.6 pp and general reasoning by +8.4 pp over the base model (better than supervised finetuning on 25 k expert trajectories).

* Synergy across games: Adding Tic-Tac-Toe (spatial) and Simple Negotiation (strategic) produces an average +9.2 pp jump on eight reasoning benchmarks; the same regimen still helps a much stronger 7-B model (+2 pp).

* Domain-agnostic data generation: Zero-sum games produce unlimited, automatically-scored training data (no problem-answer pairs or reward engineering needed).

* Scales with model strength: A model that already beats most open LLMs on math still gains from SPIRAL, hinting the approach isn’t just a crutch for weak models.

Paper: https://arxiv.org/abs/2506.24119

Related paper: Absolute Zero: Reinforced Self-play Reasoning with Zero Data https://arxiv.org/abs/2505.03335

Together, these papers trace a path toward fully autonomous curricula: start with self-generated solvable tasks, embed them in multi-agent settings that force continual adaptation, and you may get a recipe for open-ended reasoning growth without any handcrafted datasets at all.

Just a few years ago, the Grok vs. xAI saga would have seemed like the plot of a science fiction story.

Musk is super pissed that despite trying to reign Grok, it's still calling out his lies. Back in March, it called him a top misinformation spreader.

References:
- https://x.com/grok/status/1917905876301824364
- https://x.com/grok/status/1904798600409853957
- https://x.com/elonmusk/status/1935180620352958935

Links for 2025-07-02 [Part 1]

AI

1. Q1 AI Benchmark Results: Pro Forecasters Crush Bots https://www.metaculus.com/notebooks/38673/q1-ai-benchmarking-results/

2. FRI found that superforecasters and bio experts dramatically underestimated AI progress in virology: they often predicted it would take 5-10 years for AI to match experts on a benchmark for troubleshooting virology (VCT), but actually AIs had already reached this level. https://forecastingresearch.org/ai-enabled-biorisk

3. Bridging Offline and Online Reinforcement Learning for LLMs https://arxiv.org/abs/2506.21495

4. Inference-Time Scaling and Collective Intelligence for Frontier AI https://sakana.ai/ab-mcts/

5. Large Population Models https://www.media.mit.edu/projects/ai-lpm/overview/

6. UMA: A Family of Universal Models for Atoms https://arxiv.org/abs/2506.23971

7. Small Language Models are the Future of Agentic AI https://arxiv.org/abs/2506.02153

8. Visual Structures Helps Visual Reasoning: Addressing the Binding Problem in VLMs https://arxiv.org/abs/2506.22146

9. This connector tells the model how to ask the programmer whenever it’s confused, keeping the human in the loop. https://masonyarbrough.com/blog/ask-human

10. Radial Attention: O(nlog n) Sparse Attention with Energy Decay for Long Video Generation https://arxiv.org/abs/2506.19852

11. μnit Scaling: Simple and Scalable FP8 LLM Training https://arxiv.org/abs/2502.05967

Links for 2025-07-02 [Part 2]

Engineering

1. This mechanism shrinks when pulled https://www.youtube.com/watch?app=desktop&v=-QTkPfq7w1A

2. A New Pyramid-Like Shape Always Lands the Same Side Up https://www.quantamagazine.org/a-new-pyramid-like-shape-always-lands-the-same-side-up-20250625/

3. Sandia turns on brain-like storage-free supercomputer https://blocksandfiles.com/2025/06/06/sandia-turns-on-brain-like-storage-free-supercomputer/

4. Denmark is testing a fleet of robotic sailboats (sailboat drones). They’re intended for surveillance in the North Sea. https://apnews.com/article/denmark-robot-sailboats-baltic-sea-bfa31c98cf7c93320115c0ad0e6908c5

5. MIT engineers develop electrochemical sensors for cheap, disposable diagnostics https://news.mit.edu/2025/mit-engineers-develop-electrochemical-sensors-cheap-disposable-diagnostics-0701

6. New imaging technique reconstructs the shapes of hidden objects https://news.mit.edu/2025/new-imaging-technique-reconstructs-hidden-object-shapes-0701

7. Ultrafast J-resolved magnetic resonance spectroscopic imaging for high-resolution metabolic brain imaging https://www.nature.com/articles/s41551-025-01418-4

The Nature of Reality

1. P-Zombies Would Report Qualia https://www.astralcodexten.com/p/p-zombies-would-report-qualia

2. On Fine-Tuning, Bayesian Theism, and a Humble Request for a Well-Defined Sigma Algebra. https://maximumeffort.substack.com/p/highlights-from-the-comments-on-a

3. Tegmark and the Engines of Mathematics https://onemanynone.substack.com/p/tegmark-and-the-engines-of-mathematics

4. Paradigms for computation https://www.lesswrong.com/posts/APP8cbeDaqhGjqH8X/paradigms-for-computation

5. The Impact of Early Massive Galaxy Formation on the Cosmic Microwave Background https://arxiv.org/abs/2505.04687

6. Gravitational bounce from the quantum exclusion principle https://journals.aps.org/prd/abstract/10.1103/PhysRevD.111.103537

Cognition

1. The fascinating world of cognitive genetics, and the surprises we're finding in 3,640 ancient genomes. https://pifferpilfer.substack.com/p/cognitive-genetics-through-time-surprises

2. How many people do you guess would fall victim to scammers offering to ghostwrite their novels and get them published? More than you would think. https://hardresetmedia.substack.com/p/one-nz-man-vs-pakistani-scammers

3. "our findings provide the most actuarially sound evidence to date in support of the adaptationist prediction that personality traits related to time horizon are calibrated during ontogeny in response to an individual’s risk of death" https://www.psypost.org/people-who-are-more-likely-to-die-seem-to-care-less-about-the-future/

Miscellaneous

1. Finding a billion factorials in 60 ms with SIMD https://codeforces.com/blog/entry/143279

2. WikiRadio is a website that feeds you random clips of Wikipedia audio. https://www.monkeon.co.uk/wikiradio/

3. SecureBio runs a “nucleic acid observatory” that searches wastewater for new or dangerous viruses. https://securebio.org/

4. "Military spending has large and persistent effects on output because it shifts the composition of public spending toward R&D. This boosts innovation and private investment in the medium term and increases productivity and GDP at longer horizons." https://www.aeaweb.org/articles?id=10.1257/aer.20231278

5. "For nearly every low-fertility country, the UN projects either one of two outcomes: The fertility rate will flatten, or it will rise to a number somewhere between one and two births per woman … Every part of that appears to be wrong" https://www.theatlantic.com/ideas/archive/2025/06/birth-rate-population-decline/683333/ [no paywall: https://archive.is/eWm8S]

6. In the 1952 Texas gubernatorial election, incumbent Allan Shivers ran on both the Democratic and Republican tickets, beating himself 73%-25%. https://en.wikipedia.org/wiki/1952_Texas_gubernatorial_election

How big could an “AI Manhattan Project” get?

Read more: https://epochai.substack.com/p/how-big-could-an-ai-manhattan-project

Links for 2025-07-05 [Part 1]

AI

1. Energy-Based Transformers are Scalable Learners and Thinkers — “We outscale (feed-forward) transformers while generalizing reasoning/system 2 thinking to any modality/problem without requiring verifiable rewards!” https://energy-based-transformers.github.io/

2. “Multi-agent kernel generation works. We consistently generate kernels that significantly outperform some PyTorch baselines by splitting kernel optimization into specialized roles: synthesis, compilation, correctness, reasoning, and orchestration.” https://letters.lossfunk.com/p/how-many-agents-does-it-take-to-beat

3. LLM Benchmarking Shows Capabilities Doubling Every 7 Months https://spectrum.ieee.org/llm-benchmarking-metr

4. Large Language Models Are Improving Exponentially https://spectrum.ieee.org/large-language-model-performance

5. Companies like OpenAI, Amazon and Meta have supersized their spending on artificial intelligence, with no signs of slowing down. https://www.nytimes.com/2025/06/27/technology/ai-spending-openai-amazon-meta.html [no paywall: https://archive.is/G47Ep]

6. Oracle, OpenAI Expand Stargate Deal for More US Data Centers https://www.bloomberg.com/news/articles/2025-07-02/oracle-openai-ink-stargate-deal-for-4-5-gigawatts-of-us-data-center-power [no paywall: https://archive.is/QgeBR]

7. US Lifts Chip Design Export Curbs as Part of New China Deal https://www.bloomberg.com/news/articles/2025-07-03/siemens-says-us-has-rescinded-chip-software-curbs-on-china [no paywall: https://archive.is/kYTb0]

8. A deep critique of AI 2027’s bad timeline models https://www.lesswrong.com/posts/PAYfmG2aRbdb74mEp/a-deep-critique-of-ai-2027-s-bad-timeline-models

9. CEOs Start Saying the Quiet Part Out Loud: AI Will Wipe Out Jobs https://www.wsj.com/tech/ai/ai-white-collar-job-loss-b9856259 [no paywall: https://archive.is/eSDJN]

10. “The demise of the English paper will end a long intellectual tradition, but it’s also an opportunity to reëxamine the purpose of higher education.” https://www.newyorker.com/magazine/2025/07/07/the-end-of-the-english-paper [no paywall: https://archive.is/Iw30U]

11. Universities across the country are scrambling to understand the implications of generative A.I.’s transformation of technology. https://www.nytimes.com/2025/06/30/technology/computer-science-education-ai.html [no paywall: https://archive.is/nEuHp]

12. H&M releases AI model images https://www.businessoffashion.com/news/technology/hm-releases-first-images-with-ai-digital-twins/

13. “Biomni accelerates biomedical discoveries by 100x with Claude” https://www.anthropic.com/customers/biomni

14. Models often silently correct errors within their reasoning steps. They may produce the correct final answer by reasoning steps that are not verbalised, while the steps they do verbalise remain flawed, creating an illusion of transparency. https://www.alphaxiv.org/abs/2025.02

15. This was a really interesting read. Gemini 2.5 Flash is ruthless when it is rational to be ruthless. GPT-4o-mini in contrast is cooperative and forgiving, increasingly so even as that behavior becomes more dangerous. https://arxiv.org/abs/2507.02618

16. Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning https://arxiv.org/abs/2507.00432

17. Mathematicians interact with AI, July 2025 update https://www.math.columbia.edu/~woit/wordpress/?p=15039

18. Touch begins where vision ends: Generalizable policies for contact-rich manipulation https://vitalprecise.github.io/

19. The World's First AI-Native UGC Game Engine Powered by Real-Time World Model https://blog.dynamicslab.ai/

20. A foundation model to predict and capture human cognition https://www.nature.com/articles/s41586-025-09215-4

21. LeanConjecturer: Automatic Generation of Mathematical Conjectures for Theorem Proving https://arxiv.org/abs/2506.22005v1

22. KERAP: A Knowledge-Enhanced Reasoning Approach for Accurate Zero-shot Diagnosis Prediction Using Multi-agent LLMs https://arxiv.org/abs/2507.02773