Leveling-Down Justice Warrior

@viemccoy We should absolutely not slow down, that will lead to guaranteed bad outcome. Like do you think it will lead to meaningful AI safety research or establishment of UBI infra? No! It will give all the existing powerful and rich people time to react and coordinate!!

I'm usually the token optimist in any room, but I must admit I'm fairly worried right now. It feels like we are at a crossroads where we can either take a deep breath and figure out what the next optimal move is, or keep barrelling forward hoping for the best.

@xuanalogue This problem would be especially serious if the current AI paradigm turned out to be always jagged, this will likely be the case it could become a perfect substitute for humans' usefulness while remaining unable to develop its own goals/values/judgments/utility function etc

@xuanalogue I think the AI-safety people greatly underestimate the risk of bad thing caused by the combination of existing societal structure and strong AI system while greatly overestimates the risk of AI itself.

yikes

@tenobrus I personally find anti-yudkowdky view, alignment by default, to be transparently true. When you pour in enough compute the AI system will reason itself into goodness.

@tenobrus I think this greatly underestimate the risk of bad thing caused by the combination of existing societal structure and strong AI system while greatly overestimates the risk of AI itself.

are we really on the verge of unaligned singularity and mass extinction or did u just skip breakfast

@tszzl I think this greatly underestimate the risk of bad thing caused by the combination of existing societal structure and strong AI system while greatly overestimates the risk of AI itself.

if we could coordinate a global capabilities slowdown today i would likely press that magic button

@tenobrus I made the prediction that current stem culture and institution will just collapse years ago. I will update negatively about the future of humanity if it doesn't collapse

i don't think they'll really get the credit. what's going to happen is very similar to what happened with software engineering and open source. more bifurcation, new entry to the mathematical community will become impossible. existing tenured professors, terry tao, etc will be able to make massive progress and will properly disclose ai use as such, and no one will really care that they're using ai. there will be a deluge of new results from amateurs and existing grad students / low tier post docs who are suddenly producing massive amounts of output that they are clearly barely capable of understanding. everyone will have common knowledge that this is ai, regardless of disclosure, and so it will have no bearing on things like their standing in the community and public credit. there will be more news articles, but they will increasingly focus on the "ai" aspect, and no random undergrads will be able to pass off a shocking proof as their own even to the media. math publication venues will slam shut. as the job changes so dramatically and all prior signals become ~impossible to validate, advancement will be much more based on social proof and "work trials" / close in person collaboration rather than mathematical writing and output. this is just speculation, i'm not a mathematician, but this is basically the path software has taken. in mid 2026 releasing a large and well architected open source project says literally nothing about your skill as an engineer, PRs are closed to new contributors on the vast majority of existing projects, and hiring entry level engineers still happens but via very different avenues.

My paper lies on the more mathematical side of economics. I derived a representation theorem of egalitarianism that derives the classical Lorenz curve ordering using only standard decision-theoretic rationality. However, the underlying idea itself is not new, it was developed by many economists from Kolm to Atikinson. In fact, an economist Aaberge had already attempted such a result in a series of paper culminated 'Axiomatic Characterization of the Gini Coefficient and Lorenz Curve Orderings', but unfortunately his proof is incorrect. In fact, the preliminary discussion in his paper is quite interesting, as it provides a philosophically intuitive explanation and justification for the technical development that follows. This may be a little confusing if you do not already understand the purpose of such a proof. Indeed, much of the preliminary part of my own paper is devoted to providing such motivation by illustrating the new conceptual framework that this approach opens up. For more background, I recommend Matthew D. Adler's Measuring Social Welfare: An Introduction. The book is almost perfectly tailored for philosophers, as Adler translates the formal language into the language of political philosophy. I would also recommend McCarthy's papers 'Probability in Ethics', 'The Priority View', and 'The Distributive Equality', as well as John Broome's 'Weighing Goods' and 'Weighing Lives' Welfare economics is historically synomous with social choice theory, studies how information about individual preferences or well-being can be aggregated into social evaluation under various normative constraints. The field formulates these questions using the language of set theory, combinatorics. The theory of inequality measurement by contrast borrows heavily from functional analysis. It studies everything formulated in the language of analysis, under which one distribution can be said to be more equal than another, and under which qualitative orderings admit numerical representations. As for political philosophy itself, I assume you are already sufficiently familiar with its central questions. Historically, these traditions have developed largely independently and remain insufficiently integrated. I believe, however, that this separation characterized by the tools they employed is neither necessary nor desirable. They are all concerned with the same normative phenomena, but approach them using different conceptual and mathematical tools. What, then, is the major benefit of adopting such a unified framework? Besides providing the rigor of mathematics proper, it also offers a far more expressive language. Whatever one wishes to say about equality, efficiency, trade-offs, exploitation, economic planning can be formulated with precision and without ambiguity, rather than in the often imprecise language traditionally employed by philosophers and social theorists.

@irrepldjw @ProfBZZZ Links? Dm if better

@enzoreds @ProfBZZZ I had seen them decade ago and don't find them....You might be interested in my work and all the others comtemporary works from Microeconomics that study this topic

@enzoreds @ProfBZZZ My computer broke last night, let me write some paragraphs to explain

@enzoreds @ProfBZZZ arxiv.org/abs/2503.00039

@irrepldjw @ProfBZZZ See the Critique of the Gotha Programme on equality and fair distribution / equal right

Correct. Academia is dominated by left liberals (much more “I’m with her” than “Bernie would’ve won”), with a loud and politically toothless contingent of radlibs they keep as exotic pets. There are hardly any Marxists or other actual leftists.

Sorry It's Marx who is incoherent. I mean, it's not difficult to understand why Marxism has been so much more influential than others. Instead of hopelessly arguing for equality and treating people as you want to be treated in a immoral world. it is better to muddy the waters by... whatever Marxism is.

@ProfBZZZ Marx famously argued inequality was an incoherent ideal. He wanted (positive) freedom. If ideologies are a moving target then so is the median voter—you just have to ask who funds the media voters consume. These complains from Leiden—home of Cliteur & co—are odd anyhow.

@ilex_ulmus Ethics aside, this is just actively creating hostile AI as long as it's some kind of minimal entity with intention and game theoretic utility function.

@irrepldjw If you are thinking about how to appease it after it wins then you've already given up. It's never going to be easier to fight than now, and we will never be safe if we lose control to it.

Many people can't believe that AI models don't already come with kill switches, but they don't, as the OpenAI-model-escaping-containment-and-attacking-Huggingface incident showed. Contact your representatives to support this bill to require a kill switch for AI models.

@teortaxesTex The 20 billion yuan he personally invested in this round to avoid diluting his stake would had much greater impact last year. Though the number is not officically confirmed.

@irrepldjw why do you think he was in a good position financially it's only now turning to neutral cash flow High-Flyer also didn't make a killing Alibaba and BD etc. can hoover up lots of compute even damn Minimax probably had more cash

on compute, Wenfeng is extremely clear and unambiguous. Hear this? *Everything* GB300 can do, 950DT can do too; at effective 1:4 exchange rate. Fable, Mythos… "Latency and all the same". 1 full SuperPoD (8192 NPUs) = 2048 GB300s, ≈ 28 NVL72 GB300 racks, ≈3.8 MW of power.

@teortaxesTex I mean deepseek was in a much better position last year, but he didn't use that to buy compute as much as he could.

@irrepldjw he's entirely scale-pilled, wtf? He says he trains the biggest models he could why do you think he could buy anything the chinese are stingy and he's not a particularly rich actor

@ilex_ulmus Don't agree with this X-risk course, my P(doom) << 1%. But it's indeed ridiculous to think you could do any good after you earned some earthly success or "resources". That's not how humans work at all.

@andersonbcdefg I feel the same

something is still off w/ codex usage consumption. i used to be able to code and code and code and never even get close to limit. now i can use a week's limit in a day. i am using sol on medium. what is going on