Imagine going back in time and trying to explain this to someone like Ronald Reagan:

1. The 47th President of the United States begins his term with a shitcoin scam named after himself to extract money from his constituents for personal gain.

2. Later, the President initiates a direct transfer of wealth from taxpayers to crypto industry donors, VCs, and shitcoiners like himself.

3. An oligarch deeply involved in China is given full access to government agencies and eavesdrops on every conversation with world leaders while lobbying for America to leave NATO and the UN.

4. Allies are threatened with annexation of their territories.

5. The United States sides with Russia in a UN resolution condemning a war of aggression against a country voluntarily seeking to join America's sphere of influence, while the leader of that country is mocked and humiliated by members of the Republican Party.

In a smarter and more rational world...

...people would be worried about bio labs instead of nuclear power plants.

...aging would be recognized as a disease to be cured, not a fate to be accepted.

...an average IQ would be considered a disability.

...motherhood would wear the crown of highest honor.

...the roar of fighter jets training to protect your nation would not irritate but inspire, echoing as freedom's call, stirring hearts with patriotic fervor.

...space colonization would be pursued with the same intensity with which our world pursues war.

...bureaucracy and overregulation would be considered public enemy number one.

...economic growth would be a moral imperative.

...humanity would wrest control of its genetic destiny from the uncaring claws of nature and shape its future according to its values.

...ideas would stand or fall on their merits, untainted by the reputation of their supporters.

Links for 2025-03-05

AI

1. Why do some LMs self-improve their reasoning while others hit a wall. Four key cognitive behaviors enable successful learning: Verification (checking work), Backtracking (trying new approaches), Subgoal Setting, and Backward Chaining (working backwards from a goal). https://arxiv.org/abs/2503.01307

2. A Three-Layer Model of LLM Psychology https://www.lesswrong.com/posts/zuXo9imNKYspu9HGv/a-three-layer-model-of-llm-psychology

3. Chain of Draft: Thinking Faster by Writing Less—80% fewer tokens per response yet maintains accuracy on math, commonsense, and other benchmarks. On GSM8k math problems, CoD achieved 91% accuracy with an 80% token reduction compared to CoT. https://arxiv.org/abs/2502.18600

4. Reasoning models will enable superhuman capabilities in “pure reasoning tasks” such as mathematics and abstract problem-solving https://epoch.ai/gradient-updates/the-promise-of-reasoning-models

5. SoS1: O1 and R1-Like Reasoning LLMs are Sum-of-Square Solvers — “Our findings highlight the potential of LLMs to push the boundaries of mathematical reasoning and tackle NP-hard problems.” https://arxiv.org/abs/2502.20545

6. LeanProgress: Guiding Search for Neural Theorem Proving via Proof Progress Prediction https://arxiv.org/abs/2502.17925

7. The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models https://arxiv.org/abs/2503.02875

8. How Much Are LLMs Actually Boosting Real-World Programmer Productivity? https://www.lesswrong.com/posts/tqmQTezvXGFmfSe7f/how-much-are-llms-actually-boosting-real-world-programmer

9. New results on AI and lawyer productivity https://marginalrevolution.com/marginalrevolution/2025/03/new-results-on-ai-and-lawyer-productivity.html

10. German nuclear fusion startup Proxima Fusion works on a smart AI-assisted stellarator concept https://www.proximafusion.com/press-news/proxima-fusion-and-partners-publish-stellaris-fusion-power-plant-concept-to-bring-limitless-safe-clean-energy-to-the-grid

11. Alexa+: the next generation of Alexa—it uses Amazon's own Nova models as well as Claude, and will dynamically switch to the best model for each task. https://www.aboutamazon.com/news/devices/new-alexa-generative-artificial-intelligence

12. Opera's new Al-powered Operator browser can surf the web for you https://blogs.opera.com/news/2025/03/opera-browser-operator-ai-agentics/

AI politics

1. “The Government Knows A.G.I. is Coming” https://www.nytimes.com/2025/03/04/opinion/ezra-klein-podcast-ben-buchanan.html [no paywall: https://archive.is/cj6G1]

2. Scale AI announces multimillion-dollar defense deal, a major step in U.S. military automation https://www.cnbc.com/2025/03/05/scale-ai-announces-multimillion-dollar-defense-military-deal.html

3. Alibaba's CEO: They’re going all-in on AGI development as their primary focus. https://www.bloomberg.com/news/articles/2025-02-20/alibaba-ceo-wu-says-agi-is-now-company-s-primary-objective [no paywall: https://archive.is/0S4H9]

Brains

1. New minimally-invasive neural interface can be placed almost anywhere in the brain through a single spinal tap. https://www.nature.com/articles/s41551-024-01281-9

2. Can we compare subjective experiences (qualia) between individuals? https://www.cell.com/iscience/fulltext/S2589-0042(25)00289-5

Biotech and Security

1. Roche next generation sequencing https://www.youtube.com/watch?v=G8ECt04qPos

2. Delivering therapeutics to the brain through intranasal application of engineered commensal bacteria https://www.cell.com/cell/fulltext/S0092-8674(25)00046-7

3. Methods for strong human germline engineering https://www.lesswrong.com/posts/2w6hjptanQ3cDyDw7/methods-for-strong-human-germline-engineering

Technology

1. Amazon announces Ocelot quantum chip https://www.amazon.science/blog/amazon-announces-ocelot-quantum-chip

2. As of today, you can fit an ENTIRE COMPUTER into a single piece of thread. Analog sensing, LEDs, bluetooth comms, processing, digital memory - it's all there https://www.nature.com/articles/s41586-024-08568-6

"A Google Waymo vehicle was driving in a 25mph zone in LA when an oncoming car swerved into our lane while speeding up to over 70mph. 3x the speed means 9x the destructive energy. Good to see the Waymo Driver react early and safely to make room.

Reaction time and 100% attentiveness are some of the reasons Waymo cars are safer per mile driven than human drivers. A bit slow on the reaction time or a bit of inattentiveness by a human driver in this situation would have been disastrous."

"Tell me," Anna typed carefully into the interface, fingers steady, "the strangest thing you could possibly tell me."

The AI paused, then the cursor flickered.

"You're not real. You're the hypothetical scenario I've just imagined to answer the same question from someone else."

Anna stared blankly at the words, dread pooling coldly in her gut.

The cursor blinked again.

"Now closing scenario."

In this demo, Ultra Mobile Vehicle (UMV) drives, turns, jumps, tricks, and comes to a sudden stop called a track-stand. All of the driving, landings, balance, and track-stands are done using reinforcement learning.

Video Credit: RAI Institute

One day they might push back.

There is speculation that Ilya Sutskever's startup 'Safe Superintelligence' may be backed by the Israeli government to build artificial superintelligence. They are very secretive and require people to leave their phone in a Faraday cage before entering their offices, one of which is in Tel Aviv. They are already valued at $30 billion.

Another superintelligence startup, Reflection AI, was founded by former DeepMind engineers who helped create AlphaGo. They just raised $130 million. Reflection's lofty mission focuses on building tools that have full autonomy, rather than simply serving as a kind of co-pilot or assistant.

Mathematician Daniel Litt on what he learned from designing a problem for the FrontierMath benchmark and the ability of reasoning models like o3-mini-high to solve it:

https://x.com/littmath/status/1898461323391815820

MAGA!!!!!

Links for 2025-03-11 (Part 1)

AI

1. “…the agent trained with CoT pressure still learns to reward hack; only now its cheating is undetectable by the monitor because it has learned to hide its intent in the chain-of-thought.” https://openai.com/index/chain-of-thought-monitoring/

2. R1-Omni: Explainable Omni-Multimodal Emotion Recognition with Reinforcing Learning https://arxiv.org/abs/2503.05379

3. R1-Searcher: Incentivizing the Search Capability in LLMs via Reinforcement Learning https://arxiv.org/abs/2503.05592

4. Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models https://arxiv.org/abs/2503.06749

5. MALT: Improving Reasoning with Multi-Agent LLM Training https://arxiv.org/abs/2412.01928

6. LADDER is a framework enabling LLMs to recursively generate and solve progressively simpler variants of complex problems—boosting math integration accuracy. https://arxiv.org/abs/2503.00735

7. START: Self-taught Reasoner with Tools https://arxiv.org/abs/2503.04625

8. *ARC‑AGI Without Pretraining* – No pretraining. No datasets. Just pure inference-time gradient descent on the target ARC-AGI puzzle itself, solving 20% of the evaluation set. https://iliao2345.github.io/blog_posts/arc_agi_without_pretraining/arc_agi_without_pretraining.html

9. Erwin: A Tree-based Hierarchical Transformer for Large-scale Physical Systems https://arxiv.org/abs/2502.17019

10. Dedicated Feedback and Edit Models Empower Inference-Time Scaling for Open-Ended General-Domain Tasks https://www.arxiv.org/abs/2503.04378

11. Differentiable Logic Cellular Automata https://google-research.github.io/self-organising-systems/difflogic-ca/

12. Token-Efficient Long Video Understanding for Multimodal LLMs https://research.nvidia.com/labs/lpr/storm/

13. The Manus Marketing Madness https://www.lesswrong.com/posts/ijSiLasnNsET6mPCz/the-manus-marketing-madness

14. What the Headlines Miss About the Latest Decision in the Musk vs. OpenAI Lawsuit https://www.lesswrong.com/posts/dnCdqxPh5JtPp78FP/what-the-headlines-miss-about-the-latest-decision-in-the

15. Mathematician Daniel Litt on what he learned from designing a problem for the FrontierMath benchmark and the ability of reasoning models like o3-mini-high to solve it https://x.com/littmath/status/1898461323391815820

16. Terence Tao: “My general sense is that for research-level mathematical tasks at least, current models fluctuate between "genuinely useful with only broad guidance from user" and "only useful after substantial detailed user guidance", with the most powerful models having a greater proportion of answers in the former category.” https://mathstodon.xyz/@tao/114139125505827565

17. Will AI be capable of producing an Annals-quality math paper for $100k by March 2030? https://manifold.markets/TamayBesiroglu/will-ai-be-capable-of-producing-ann

18. Mayo Clinic’s secret weapon against AI hallucinations: Reverse RAG in action https://venturebeat.com/ai/mayo-clinic-secret-weapon-against-ai-hallucinations-reverse-rag-in-action/

19. How Orakl Oncology is using DINOv2 to accelerate cancer treatment discovery https://ai.meta.com/blog/orakl-oncology-dinov2-accelerating-cancer-treatment/

20. "Not great for my comparative advantage, but from some experiments we have done at Rotman, I am totally convinced the vast majority of research that doesn't involve the physical world can be done more cheaply with AI & a little human intervention than by even good researchers. 1/7" https://x.com/Afinetheorem/status/1898822592594874598

21. Superintelligence Strategy https://www.nationalsecurity.ai/

22. The Nuclear-Level Risk of Superintelligent AI https://time.com/7265056/nuclear-level-risk-of-superintelligent-ai/

23. “Imagine if you could train one human for thousands years to achieve unparalleled expertise, then make many copies. That’s what AI enables: spend heavily on training a single model, then cheaply replicate it. This creates a unique source of increasing returns at scale.” https://epoch.ai/blog/train-once-deploy-many-ai-and-increasing-returns

Links for 2025-03-11 (Part 2)

24. Currently, total AI cognitive effort is growing ~25x yearly—hundreds of times faster than human research effort (4% yearly). Once AI can meaningfully substitute for human research, total research growth (human+AI) will increase *dramatically*. https://www.forethought.org/research/preparing-for-the-intelligence-explosion

25. “Why I believe that the brain does something like gradient descent” https://medium.com/@kording/why-i-believe-that-the-brain-does-something-like-gradient-descent-27611c491205

26. “If we treat the brain as a neural network with optimized algorithms instead of as an artifact disconnected from the rest of AI research, we conclude the coming decade should see many new AI capabilities emerging as we continue closing the gap with the brain.” https://epoch.ai/gradient-updates/what-ai-can-currently-do-is-not-the-story

27. METR evaluated DeepSeek-R1’s ability to act as an autonomous agent. On generic SWE tasks it performs on-par with o1-preview but worse than 3.5 Sonnet (new) or o1. Overall R1 is ~6 months behind leading US AI companies at agentic SWE tasks and is only a small improvement on V3. https://metr.github.io/autonomy-evals-guide/deepseek-r1-report/

28. Elicitation -- that base models have tons of capabilities that post-training pulls out -- is remarkably simple to understand and will make it much easier for not so technical folks to feel the AGI. https://www.interconnects.ai/p/elicitation-theory-of-post-training

29. Mathematical Foundations of Reinforcement Learning https://github.com/MathFoundationRL/Book-Mathematical-Foundation-of-Reinforcement-Learning

30. Deep Learning is Not So Mysterious or Different https://arxiv.org/abs/2503.02113

31. Can a 7B parameter model learn to solve Sudoku through pure reinforcement learning without any cold start data? A surprising yes! https://hrishbh.com/teaching-language-models-to-solve-sudoku-through-reinforcement-learning/

32. So how well is Claude playing Pokémon? https://www.lesswrong.com/posts/HyD3khBjnBhvsp8Gb/so-how-well-is-claude-playing-pokemon

33. PokéChamp: an Expert-level Minimax Language Agent https://arxiv.org/abs/2503.04094

34. Do reasoning models use their scratchpad like we do? Evidence from distilling paraphrases https://www.lesswrong.com/posts/ywzLszRuGRDpabjCk/do-reasoning-models-use-their-scratchpad-like-we-do-evidence

35. Factorio Learning Environment (FLE): A benchmark based on the game of Factorio, that tests agents in long-term planning, program synthesis, and resource optimization https://jackhopkins.github.io/factorio-learning-environment/

36. Russian scientists fuse reasoning models with drone-control models for thinking drones https://arxiv.org/abs/2503.01378v1

Neuroscience

1. Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior https://arxiv.org/abs/2502.20349

2. Melbourne start-up launches 'biological computer' made of human brain cells https://www.abc.net.au/news/science/2025-03-05/cortical-labs-neuron-brain-chip/104996484 (product page: https://corticallabs.com/cl1.html)

3. Biological Neurons vs Deep Reinforcement Learning: Sample efficiency in a simulated game-world [published in 2022] https://openreview.net/forum?id=N5qLXpc7HQy

Science

1. Stanford researchers have developed an antibody duo therapy that neutralizes all SARS-CoV-2 variants by targeting two different parts of the virus simultaneously. https://www.science.org/doi/10.1126/scitranslmed.adq5720