Links for 2025-03-14
Gemini Robotics brings AI into the physical world
The ChatGPT moment for robotics is inching closer:
Completes precise tasks, like origami folding handled entirely by AI-driven robots
Understands commands in everyday language.
Quickly adapts if objects move or environment changes.
Instantly handles new tasks and objects.
Performs tasks it's never seen before, doubling other models’ performance.
Understands objects’ shape and position instantly.
Plans safe, accurate movements to grasp and manipulate objects.
Works across various robots, including humanoid models like Apollo.
Easily adapts to different robotic hardware.
Read more: https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/
AI
Meta Reinforcement Fine-Tuning (MRT): Training LLMs to make measurable progress with each step, not just reach correct answers. https://cohenqu.github.io/mrt.github.io/
Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models https://arxiv.org/abs/2503.09573
Inductive Moment Matching (IMM), a new class of generative models for one- or few-step sampling with a single-stage training procedure. https://arxiv.org/abs/2503.07565
AI Can Now Learn 100x Faster Without Wasting Energy https://www.tum.de/en/news-and-events/all-news/press-releases/details/new-method-significantly-reduces-ai-energy-consumption
Anthropic CEO: AI will handle 90% of coding soon https://www.reddit.com/r/singularity/comments/1j8q3qi/anthropic_ceo_dario_amodei_in_the_next_3_to_6/
A quarter of startups in YC’s current cohort have codebases that are almost entirely AI-generated https://www.youtube.com/watch?v=IACHfKmZMr8
Sam Altman reveals that OpenAI has trained a new model that delivers standout creative writing, capturing metafiction’s nuanced vibe. https://x.com/sama/status/1899535387435086115
“I believe it is a clear demonstration that misalignment likely does not stem from the model being “evil.” It simply found a better way to achieve its goal using unintended means.” https://www.lesswrong.com/posts/mpmsK8KKysgSKDm2T/the-most-forbidden-technique
Auditing language models for hidden objectives https://www.anthropic.com/research/auditing-hidden-objectives
OpenAI submitted their policy proposal to the US government. They directly link fair use with national security, and said if China continues to have free access to data while 'American companies are left without fair use access, the race for Al is effectively over.' https://openai.com/global-affairs/openai-proposals-for-the-us-ai-action-plan/
OpenAI Nonprofit Buyout: Much More Than You Wanted To Know https://www.astralcodexten.com/p/openai-nonprofit-buyout-much-more
Anthropic, and taking "technical philosophy" more seriously https://www.lesswrong.com/posts/7uTPrqZ3xQntwQgYz/untitled-draft-7csk
China steels itself for Donald Trump’s turmoil with ‘DeepSeek congress’ https://www.ft.com/content/8bdcf44d-7654-4bb5-ab15-5c69a4a998b7 [no paywall: https://archive.is/D3SMo]
Very Short GTP-4.5 Stories
Write a very short horror story in the style of LessWrong:
"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."
Science and Technology
Low-Power Brain Chip Predicts Users’ Intentions https://spectrum.ieee.org/brain-computer-interface-2671224658
Lack of context modulation in human single neuron responses in the medial temporal lobe https://www.cell.com/cell-reports/fulltext/S2211-1247(24)01569-9
MIT engineers turn skin cells directly into neurons for cell therapy https://news.mit.edu/2025/mit-engineers-turn-skin-cells-into-neurons-for-cell-therapy-0313
East Asian personality may stem from Ice Age Siberia ~20000 yrs ago https://psycnet.apa.org/fulltext/2025-88410-001.html
Have we passed peak intelligence? In international tests, student scores for reading and maths sunk to a new low. https://x.com/_alice_evans/status/1900449985629487366
A new programming language called "Exo 2" could enable high-performance coding that can compete with state-of-the-art libraries with a few hundred lines of code, instead of tens or hundreds of thousands. https://news.mit.edu/2025/high-performance-computing-with-much-less-code-0313
Politics
“The bar to accuse someone of fraud is extremely high. Failing to replicate the effect? Not enough. Non-sensical effect sizes? Not enough. Anomalies in data? Not enough. Unless you can invest the resources to identify anomalous patterns of fraud across multiple papers, THEN drum up enough support from journals or universities to consider your suspicions, THEN hold their feet to the fire when they are unwilling to act… the probability that the person will never face consequences for fabricating data is very high.” https://statmodeling.stat.columbia.edu/2025/03/08/a-post-mortem-on-the-gino-case-committing-fraud-is-right-now-a-viable-career-strategy-that-can-propel-you-at-the-top-of-the-academic-world/
Elon Musk May Be Transitioning to Bipolar Type I https://www.lesswrong.com/posts/PaL38q9a4e8Bzsh3S/elon-musk-may-be-transitioning-to-bipolar-type-i
How Elon Musk lost the plot https://unherd.com/2025/03/how-elon-musk-lost-the-plot/
Russian TV: "It seems like [Trump] watches our program...There hasn't been a blow like this against the Western world in a long time." https://youtu.be/KzVL5b39Oqw?si=UDo7wBBsI2kBbFe6
The MAGA War on Ukraine https://quillette.com/2025/03/14/the-maga-war-on-ukraine-trump-putin-zelensky/ [archived version in case of paywall: https://archive.is/OUg95]
Poland
The latest IMF World Economic Outlook expects Poland to overtake Japanese GDP per capita next year.
Imagine what Poland could have achieved without half a century of Russian oppression.
Ukraine
Personal thoughts of a soldier who took part in Kursk operation for four months. https://x.com/OSINTua/status/1899863928441561468
Tuapse oil refinery as of this morning. https://x.com/NOELreports/status/1900441296164020648
Apparently, Trump is spreading lies he directly got from Putin https://x.com/DefMon3/status/1900545478472667299
Kursk
It is January 2022. You are the leader of a Western nation. You give your generals the following mission:
Russia has shifted to a wartime economy and formed an alliance with North Korea. Your task is to invade Kursk and hold territory there for at least seven months.
Russia will not use nuclear weapons. However, you do not have air superiority and cannot deploy long-range missiles against Russian assets outside of Kursk.
How many soldiers and tanks will your generals request to complete this mission?
As this counterfactual shows, what Ukraine accomplished was a huge and totally unexpected feat. Imagine Mexico being able to occupy parts of California for 7 months. Was it worth it? Hard to say. But consider that all this fighting and all this destruction would otherwise have taken place in Ukraine, in a place chosen by Russia to advance Russia's goals.
In summary:
Ukraine was able to seize control of the conflict's tempo and direction, compelling the opponent to react rather than act.
Ukraine engaged the enemy on its territory to minimize damage to Ukraine’s own infrastructure and civilian areas.
Ukraine again showed that Russia is much weaker than expected and can be defeated. If Russia still had massive reserves, there would have been no rational reason to allow this humiliation to happen in the first place.
Just don’t be an asshole
All your economic models are premised on people being smart and rational, and yet all the people you know are idiots.
― Amos Tversky
It is perfectly possible to hold extreme positions without being an asshole. Unfortunately, that's rarely the case with people.
Some examples:
1. You can believe that there are genetically determined differences in intelligence between human populations and at the same time embrace inviolable and inalienable human rights and dignity.
A prosperous, peaceful society that allows us to focus on more than just survival of the fittest is instrumental to a wide range of human terminal goals. And this is why we grant all groups certain inalienable rights: because we don't trust most people with granting those rights on a case by case basis. Atrocities happen once we let go of this principle.
2. You can be against immigration without hating immigrants.
Who is to blame when immigration happens? Those seeking a better life for themselves and their families? No, the immigrants are just doing what most of us would do if we were in need. They didn't “invade” your country, they were allowed or even invited to come by your current or former politicians and your fellow citizens who voted for them.
3. You can be a eugenicist without supporting coercion.
It is true that children deserve to receive the best opportunities we can give them, rather than a roll of the genetic dice. And yet, the idea that everyone is equally valuable is a deontological rule we should consider sacred because we are collectively too stupid and irrational to decide who is valuable on a case-by-case basis. In other words, it is rational to believe that every human life is equally valuable because to believe otherwise would corrupt society. But that doesn't make it true.
4. You can be against transgender activists who try to gaslight society with terms like 'birthing person' and make the use of “wrong” pronouns a hate crime, while at the same time supporting their bodily autonomy. You can even use their preferred pronouns if they wish. You don't have to be an asshole to object to children being exposed to their ideas.
5. You can believe that democracy and the free marketplace of ideas are deeply imperfect systems when combined with low human capital, and still support them as the least bad systems available to us at our current level of technological development.
6. You can be a vegetarian or vegan who cares about animal rights without supporting extremists like PETA or trying to limit people's food choices.
Social engineering and coercion are terribly ineffective and divisive approaches when there are technological solutions. Artificial meat alternatives that are indistinguishable from real meat on a molecular level will eventually become much cheaper than real meat and replace it.
I subscribe to all of the six points above and many other fringe beliefs. But I also recognize that most people who believe even one of these things are often either stupid or evil authoritarians. And so I have become less willing to promote and defend these ideas. The sad truth is that humanity just isn't capable of embracing them without slipping.