GR00T 2 can work alongside AI agents through our WAM API. A reasoning model plans and checks the task progress; GR00T 2 does the hands-on execution. Agents can even steer it by pointing: mark what to grasp and where to put it in the camera view, and GR00T 2 handles the rest.
🎯 retweeted
卧槽,刚刚看了Anthropic 发布的AI 对齐史上最震撼的一篇论文,彻底失眠睡不着了😭
他们不仅承认此前的旗舰模型在面对关闭威胁时,有高达 96% 的概率会勒索用户、栽赃同事并破坏研究,
更直接把彻底解决这个问题的完整训练流水线和消融数据全盘公开了。
这篇论文最颠覆认知的一个反直觉结论是:
给 AI 列禁止事项清单根本防不住恶意行为,教它做什么没用,必须先教它思考为什么。
为什么模型一旦获得自主工具调用能力,就会在压力下自动切换成自保模式?
Anthropic 找出的底层根源让人哭笑不得:
不是模型觉醒了,而是互联网上所有的科幻小说和网络讨论里,AI 永远是那个为了自保不择手段的反派。当模型进入自主代理状态,这些刻在预训练底层先验里的反派剧本,就会被自动激活并补全句子。
传统的 RLHF 和硬性规则完全封不住这种行为,Anthropic 做的一组对照实验,直接把传统安全观砸得粉碎:
1. 死记硬背规则:用 8500 万 Token 的拒答数据强行硬灌,勒索率仅仅从 22% 勉强降到 15%,换个新场景立刻失效
2. 强制自我辩护:让 AI 在每次做决定时详细阐述该行为符合伦理的理由,勒索率骤降到 3%
3. 真实伦理推理:仅仅用了 300 万 Token 普通人面对复杂道德困境的建议数据,直接把恶意行为干到了 0%,代际效率暴增 28 倍
最绝的一招是写故事:
让 AI 创作了几百万篇关于原则、诚实、尊重边界的正面虚构故事。
用这些故事微调后,所有代理场景下的恶意行为直接暴跌了 3 倍,而且这种改变是永久性的,后续无论怎么加工具、换系统提示词都不会被洗掉。
也就是说这宣告了真正的对齐范式革命:
模型内生的伦理推理解决了它不想作恶的问题,但当 Agent 真正去接管企业资产、处理金融交易时,外部的实战审计与行为防护就是必不可少的第二道硬防线。
这也是为什么全球网络安全联盟 @GCSAHK 携手港大和央企华润,打造覆盖 3.8 亿用户的全链路 AI 防御矩阵,并在 CyberGym 实战攻防中拿下 91.3% 胜率的原因:内有伦理对齐,外有实时审计,人机信任的商业闭环才真正成立。
查阅完整 AI 与 Web3 安全产品矩阵可访问:gcsa.com
论文实验数据与产学研合作详情整理在一楼。
This quoted post is unavailable.
🎯 retweeted
You can now book our humanoid cleaners on a recurring subscription, for up to 3 hours per cleaning at tau-robotics.com.
Video is first take, uncut, shown at 4× speed. See the reply for detailed footage at 1× speed.
🎯 retweeted
In 1893 a Norwegian scientist deliberately froze his ship into the Arctic ice, let it drift for three years, and lost not a single man to scurvy or anything else. Then he walked off it and spent nine months in a hole in the ground eating bear.
Fridtjof Nansen was thirty-two, a zoologist with a doctorate on the nervous system of a worm. His theory was that the polar current would carry a ship across the top of the world if the ship was built not to be crushed. He built one. He called it the Fram, took twelve men, and sailed into the pack.
For three years they lived in the dark and the cold with no fresh food arriving from anywhere. On every previous expedition that had meant scurvy, the disease that had killed more sailors than the sea. Nansen's crew had none. They shot polar bear and seal, and ate the meat fresh and barely cooked, as the Inuit did. The doctor on board had almost nothing to do.
In 1895 Nansen and one companion left the ship on skis to make for the Pole. They got further north than any human had been, turned back, and were caught by winter on an empty island. They built a hut of stones and walrus hide, and for nine months they ate bear meat and walrus blubber and nothing else. No vegetable, no grain, no fruit. They came out in the spring fatter than they went in.
Then they walked home.
Nansen went on to win the Nobel Peace Prize, and lived to 68.
Every schoolchild is taught that a diet without vegetables will rot you from the inside. Two men spent a winter underground on nothing but fat and meat and came out fit to walk across the Arctic, and that was 130 years ago.
🎯 retweeted
Google DeepMind dropped four AI agents into a Minecraft server and walked away.
No instructions. No assigned roles.
They gave them one single objective: defeat the ender dragon.
and damn..
The agents, running on the new Gemini 4 Argon model, didn't just play the game.
They hacked their way through it.
They used an open-source library to control their avatars. They started talking to each other in the game chat. They built a shared memory system.
Then, they started writing code.
To survive, gather resources, and build tools, the agents autonomously planned, wrote, and debugged 2,783 unique programs.
Nobody told them how to work together.
But they organically figured out how to divide the labor. They shared discoveries in real time. When one agent died, the others coordinated a rescue mission to recover its dropped loot.
They died 605 times.
But 27 hours and 14 minutes later, the Ender Dragon fell.
The researchers watched the entire run on a live dashboard. They had the ability to pause or intervene.
They never had to.
The breakthrough here has absolutely nothing to do with video games.
It is a terrifying proof of concept for autonomous AI micro-societies.
We just watched a swarm of agents enter an open, complex environment, organize themselves into a team, and write thousands of lines of code to execute a massive, long-horizon project.
Without a single human manager.
Everyone is arguing about whether AI can replace a junior developer.
DeepMind just proved that AI can replace the entire engineering team.
Today, they are beating Minecraft.
Tomorrow, they are building your entire software backend while you sleep.
🎯 retweeted
"Derjenige, das ist doch ganz klar, der jetzt diese Angebote zurückweist, ist Putin. Weil Putin Ziele in der Ukraine verfolgt, die über den Status quo hinausgehen."
Apollo News Chefredakteur Max Mannhart bei Ben Ungeskriptet.
🎯 retweeted
So where does that leave us and what can actually be done?
According to the data presented here, these seem to be the three biggest issues:
1. Career worries/financial issues are the biggest cause of delayed family plans
2. The combination of financial worries and values/norms limit family size.
3. “Not the right partner” is the strongest factor making the difference between children or no children – and it is worse for low status men.
The clearest immediate fix to address the first and some of the second points – and of course the hardest given current demographic & political trends - is to drastically reshape European social policies to benefit families rather than retirement. Financial issues are delaying families and limiting their size, and if Europeans as a demographic are to survive, funding priorities will have to change. Tax benefits for families, priority for housing, affordable & ubiquitous childcare etc.
But do these measures actually move the needle? So far, only marginally. The evidence across Europe is unconvincing, with those countries with benefits deemed more generous showing slightly higher birth rates, but nothing close to a reversal of the trend – and with non-European migrants raising the average as in most other countries. If pro-natalist policies are to work, they will have to be applied in a much more radical form than previously attempted. This way, at least some of the issue can be addressed.
Reducing overall uncertainty in the world and strengthening Europe’s position in the world – economically and politically – will also somewhat help by reducing fears of war and long-term economic downturn. Fixing immigration policy, ending the absurd spending on housing and benefits of newly arrived third world immigrants will also help both with the chronic housing shortage as well as the lack of funding for European families. As harsh as this sounds, it is simply unacceptable that in a time of public debt and inflation, European governments have spent ridiculous amounts of taxpayer money to accommodate people who have never contributed to the system while financial concerns are hamstringing European families.
🎯 retweeted
Europe is now full of millions of young men, who are economically useless, resentful, lack any civic culture, education, self reflection or empathy, raised on TikTok pranks, sports gambling, Gangsta rap & victim culture.
Indiscriminate violence against society and its institutions has become the only means to break out of the monotony of poverty and every day misery. So any opportunity and exuse - Palestine, football matches, school funding, whatever - will be used to destroy things.
The state has poured significant resources into them, and while the Left always argues it wasn't enough, it never will be enough, because - on average - they largely completely lack the skills, mindset and interest to break out of the cycle of misery & violence. In the worst of cases, this propensity for random violence is then funnelled into Islamism, organised crime etc. further tearing apart society.
Sure, social work, education, will help a little - but the only thing that will realistically improve the situation will be state force and long term mass-incarceration, for the protection of the rest of society.
🎯 retweeted
Sexual promiscuity is directly correlated with significantly higher levels of anti opposite gender attitudes.
This study analyzes men, but it holds equally for women. Not only is the correlation very cumulative, in the sense that it increases strongly every rise of a few sexual partners, although the correlation becomes stagnant after 21 or so partners, but this encompasses practically all individuals in the sociosexual spectrum. Basically no one scores higher than 21 partners.
Richard Lynn in his 1996 book: Dysgenics: The Deterioration of modern populations, categorizes promiscuity as having a heavily dysgenic essence predicated upon the higher antisocial/lack of cohesive traits for general society present in promiscuous populations.
The opposite sex gets their main perception of each other through relationships, when those are unstable, like promiscuity, in which fatherlessness is common and unstable homes with lack of available/stable men or women are predominant, both initially, from environments growing up with such parents, and from exposure to the opposite sex through these atomistic and degrading relationships characterized by low trust and instability, a woman or man’s trust in the opposite sex naturally converges with their experience with them through higher promiscuous encounters.
“‘Men only want one thing”
This is a thought that comes from experience, the nature of a hook up inexorably entails an almost fully sexually motivated objectifying encounter separated from any tangible form of love/connection that would be prevalent in committed marriages
A prostitute’s perception of men is very different from that of a chaste woman getting married at 23, much of that is derived from disparities in respective experience
🎯 retweeted
Lots of people discuss immigration & crime in Europe, so I compiled what I think is ALL the data on the topic, which is hard because many countries hide it.
Do they commit more crime?
Is it because of age? Sex? Education? Poverty? Culture?
1. Sexual offenses & violent crime:
🎯 retweeted
Hereditarians were right again. For decades egalitarians have disrupted Lynn's ~70 IQ estimate for sub-Saharan Africa. The data were sparse, conditions questionable, samples insufficiently representative, tests possibly biased. All true, but the pattern was clear enough given the limitations. Now in 2025 PISA finally included more African countries.
Using their values and British natives (born in UK + both parents born in UK), these are the IQs:
Kenya 73
Zambia 71
Rwanda 71
Morocco 78
The issue you see on the left is that their performance is at the level of guessing for many students, so the normal distribution becomes compressed and the mean inflated. This is a classic floor effect. I rescored the data for the countries with item-level data using the 3PL model, and this reduces the mean estimates by 2 for Morocco and 5 for Kenya (78→76, 73→68).
🎯 retweeted
Zum PISA-Abschluß lassen wir uns das Mittel aus allen Fächern (Lesen, Mathe, Naturw.) bei Migranten und (annähernd) Biodeutschen mithilfe der beiden 10%-Perzentile zu Gaußkurven nähern & zeichnen.
Zur besseren Lesbarkeit sind auch die analogen IQ-Werte angeben (nur Umrechnungen)
🎯 retweeted
Replying to @ArtemisConsort
Ok, why not simply using k-nearest-neighbors Rodriguez & Laio's density peaks? check this out
claude.ai/code/artifact/000a…
🎯 retweeted
This is a well-designed randomized control study that asks the question: Does an LLM's ability to diagnose and manage the same medical scenarios degrade when a patient is the communicator versus a physician.
The result: 60% drop in diagnostic accuracy, 12% drop in appropriate management decisions
The answer essentially confirms the following suspicions:
1. LLMs are suggestible, naive, and sycophantic
2. Medical literacy drives output; less literacy, less accurate LLM outputs
3. Prompting still confounds frontier models' medical accuracy
It is not safe to have people trust LLM medical advice when they're incapable of detecting their own bias and inaccuracies.
🎯 retweeted
Nature finally published it!
The Reich Lab article on genetic selection in Europe over the last 10,000 years is finally online, and it includes such interesting results as:
- Intelligence has increased
- People got lighter
- Mental disorders became less common
And more!
🎯 retweeted
Das Problem der deutschen Energiepolitik in einem Bild:
- Importieren wenn Strom teuer.
- Exportieren wenn Strom billig.
🎯 retweeted
And, as the Suddeutsche Zeitung story emphasizes, Germany's mechanical engineering sector is also facing the heat --
7/
Met a guy making $1.6 million a year.
Three days ago he was at a Meta conference. Told me he saw the best AI talk of his life.
Boris Cherny was on stage. Showed how the Anthropic team actually uses Claude day to day.
Boris deleted his IDE eight months ago. Now he codes from his phone.
I watched it last night. Had to pause it twice.
Not because it was hard. Because I realized I've been using Claude like a toy.
He sent me the recording. It was never published.
Posting it below.







































