Dunno, about 10% of US adults are millionaires but only about 9% are union members.

We need to run a hardcore neoliberal so y'all can finally recalibrate where the center is.
The last three candidates were centrists yes, right in the middle of Dem voters' preferences.

"Leftists can't even win over the dems but they can win the nation" is a bold take.

The Renters' Plights Bill The Renters' Rights Bill will harm private renters more than it helps them

open.substack.com/pub/benhopki...

Labour is not a serious party, plain and simple.
Issue with Britain is neither are any of the other parties.

Let me be clear: if I get citizenship before the next GE at this rate I am open to voting Reform simply because a gamble is better than a sure loss.

I lol'd

I've got Monday blues, can we immanentize the eschaton already?

I'm probably not bitter-pilled enough but I just don't believe in autoregressive language modelling anymore.

I'm gonna have a deeper look into diffusion language models and watch the mainstream pass me by.

I don't want to be racist against British people but there are entire subtypes of them that I can't tell apart.

Like I almost went "hey fancy seeing you here" to a friend of my flatmate's" before realising he was not the same guy that I met 3 weeks ago.

Every time I dig into an LLM paper I can feel myself getting crankier.

Everything is terrible: the ideas, the implementation, the reasoning, the frameworks, the hardware...

There's like 3 good things in there and the rest is slop built on top.

I'm like 5 papers and 3 repo away from vixra, max.

If you're really into tantrism it's pretty normal, you might even wonder why they didn't start with that.

Now then this begs a question: why *is* there qualia then, if it is one-way? Why does the universe have a screen?

I have no answer to that, it is for me the essence of the hardness of the problem. But it feels like a coherent way of thinking at least.

a world without qualia generates *exactly the same discourse about qualia* as a world without.

It is, in a sense, the universe's sense of humour that qualia is both apparently-universal and utterly incommunicable, because you'd utter exactly the same statements whether you have qualia or not.

but still independent of *whether the screen is plugged in* (in fact, most video games work this way: they'll generate a frame object and bug out if there's a problem in the process, but care little if the screen is plugged in).

This, as I understand it, is the intuition behind p-zombies:

all the computation necessary for the simulation itself is done without reference to the frames being displayed on the screen, and yet the frames and the screen exist. So then with consciousness.
And in fact you could create a simulation that is aware of the *frames*...

To the point where I'm wondering if I'm reading it right. My inclinations are pretty straightforwardly epiphenomenal, and I don't recognise it at all in that description.
An illustration would be video games, or perhaps even better a simulation that is recorded and displayed:

reading this because I can't sleep and it feels... hopelessly tangled? footnotes2plato.com/wp-content/u...

mostly because at a first and sleep deprived reading I found exactly no value or insight in it, and I'm glad gwern apparently feels the same way.

or as a refutation of them. I assume the latter since it criticises the list of lethalities.
If the former, then the entire blog requires some deeper x-risk context savviness than mine, at which point the audience is literally just Dwarkesh.
If the latter, then I'm going to facepalm very loudly.

"Deep Learning" Is Function Approximation — LessWrong A Surprising Development in the Study of Multi-layer Parameterized Graphical Function Approximators As a programmer and epistemology enthusiast, I've…

in the links criticising yudkowsky (always a good bellwether: if someone accuses Eliezer of being fundamentally mistaken, my money is always on Eliezer), it links this: www.lesswrong.com/posts/Dhjcdz...

Now I don't even know if it linked as an example of Eliezer's bad categories,

On Those Undefeatable Arguments for AI Doom <h1>Part 1</h1> <p>Suppose you're a student at the Catholic Franciscan University of Steubenville, who is -- somehow -- unfamiliar with the contents of Catholicism.</p> <p>One day over lunch, it comes...

I.e. there are (at least) two issues here:
- disagreements that occupy insiders may not be particularly fundamental
- even if they *are* fundamental, convergence in goals is not convergence in epistemological validity

There are other things I found weird, like...

On Those Undefeatable Arguments for AI Doom <h1>Part 1</h1> <p>Suppose you're a student at the Catholic Franciscan University of Steubenville, who is -- somehow -- unfamiliar with the contents of Catholicism.</p> <p>One day over lunch, it comes...

The catholic parable 1a3orn.com/sub/essays-a... is a case in point. Maybe I just don't get it, but I can't help but add to it "and then a solipsist came around and said "you believers in objective reality can't even agree on the validity of proofs of God?!""

Dwarkesh posted positively about this blog: 1a3orn.com and...
I just don't get it?
Like part of me wonders if anti-AI-safety posters have become the serious conservative intellectuals of the 2010s: so vanishingly rare that people would drop their standards and take yarvin seriously.

Interesting thing about their page 5 table: the biggest drop is from 1-step gradient.
Which makes sense if I think of their process decoupling reasoning from specifics, since reasoning is a deeply composable process and you really want to learn that.
But not sure about my interpretation.

The TRM paper arxiv.org/pdf/2510.04871 is a joy to read even if just for the (gentle and respectful) dunk on HRM in the targets for improvement section.

This post reminded me that I need to stop fucking around on WandB and catch up on this week's arxiv haul.

Why am I so much more polite to Claude than to ChatGPT?

This is why glazing happens I swear to God. You never realize how easily manipulated you are until you start switching personal assistants regularly.

The fun thing about running experiments on my desktop gpu via ssh is that it looks like my computer is freaking out for no reason. GPU at full speed, nothing visible on desktop.

The amount of LLM bullshit I see on this app. Not bullshit written by LLMs, bullshit about LLMs written by humans.
Latest culprit: "LLMs can't write summaries".
Bitch this ain't 2023 anymore.

Pretty much my thoughts yeah.

Looking back, the shortest non-modular reactor build that I know of was Vallecitos, at 21 months. Dubiously applicable to today but where there's a will there's apparently a surprisingly fast way.

My instinct is to write Chinese names the Chinese way. So for the authors of Physics of LLMs I write Li Yuanzhi for Yuanzhi Li, but for Zheyuan Allen-Zhu my brain hits a bug and starts writing Zhu Zheyuan Allen.

nuclear power can also take 3 years, it's really a matter of policy.
The plants are easy to build, it's the legal environment that's hazardous.