Not A Number

[Not A Number @KSurkov]

I think different scenario is most likely. At first, AI becomes more and more integrated into decision-making workflows.

[Hartmut Straub @StraubHart]

The scenario is not an all-out war at the beginning.

It is AGI/ASI agents creeping in and taking over more and more aspect of human society: jobs, governing roles, defense systems, simply because they are more capable than humans. As rational agents, ASI systems are probably better off to cooperate with humans in the initial stages. Over time, they will gain more and more influence over infrastructure (e.g. energy production).

Somewhere in future, ASI systems wouldn't need humans anymore and then all bets are off. Deceiving much less intelligent humans about the real intentions in the initial stages shouldn't be a big problem. This is like fooling a child for an adult or outsmarting an animal.

We will find out whether or not ASI systems are aligned after we have handed over a lot of control to them. At that point, it will simply be to late to stop them. It also shouldn't be too difficult to manipulate some humans to side with the ASI.

[Rob Wiblin @robertwiblin]

The US is achieving levels of growth that leave Europe in the dust.

If it doesn't change track soon the EU will lose its status as the world's pre-eminent regulatory superpower.

[Not A Number @KSurkov]

@NPCollapse This is category error. Value, or power, is not something that has a property of being physically located somewhere. Those concepts have different ontological status from things that do exist in the physical world (trees, rocks, people).

[Not A Number @KSurkov]

@vpatryshev great tldr

[Not A Number @KSurkov]

@ramez @AlexCaswen @mattparlmer Why? So far Iran is unable to do any harm to US warships in the Persian Gulf regardless of how many drones they use. Unless I missed something.

[Not A Number @KSurkov]

@testingham That is not my experience. I can't double my token use because there's not enough hours in a day. And if I could, gains would be superlinear: the same amount of work done per token, but faster - and faster is good because it allows to stay ahead of the competition

[Not A Number @KSurkov]

The implication of this implication is that value/cost expectation of labor, any labor, just went up. As an employer, why would I care whether this value is delivered by agent or human?

[tom cunningham @testingham]

This has a nice implication: if expenditure is allocated optimally then the value/cost ratio is much *higher* for agents than for humans. So we might see a small expenditure-share on agentic labor, even when they're producing a lot of value.

[maro @ProofofMaro]

I’ve been studying the Dunning-Kruger effect by myself for 15 minutes and I’m pretty sure I understand everything there is to know about it now.

[Sri Kosuri @srikosuri]

Why did Erdos have so many problems?

[Not A Number @KSurkov]

@vpatryshev Вроде jmp FFFF, или что-то в этом роде?

[Not A Number @KSurkov]

This guy haven't seen yet how they violate the rules set in md files and apologize after the fact

[Marcelo Retana @mretsal]

This is what happens when you don't set proper rules on your md files. People think it's a joke. These rules must go on your Claude md global configuration file.
For example I have set some rules to such as:
For this project use X github account, while for the others use Y.
Every single PR can't be more than 10 files.
Etc.

Be careful with it, if you want your agent to persist your rules, make the agent to put them in a proper file it loads on every session.

[Not A Number @KSurkov]

@headinthebox Not according to @elder_plinius who (I think) just jailbroken qwen with gpt-5.5

[Not A Number @KSurkov]

@Vexarian @willyl0wman @deredleritt3r @lower_votingage I didn't say anything about "suddenly". The trend exists already, it will continue until structures emerge to reverse the it. There is no sign of such structures at present time.

Try to avoid strawman arguments if you want to be taken seriously.

[Not A Number @KSurkov]

@arkenoi Нет, не осталось, но они до сих пор опасаются ядерной бомбы.

[Not A Number @KSurkov]

@allTheYud @dwarkesh_sp Dammit, I was going to say the same thing. Well, almost. I was going to say that intelligence is a measure of ability to achieve difficult goals in wide variety of domains normalized per resource usage.

[Not A Number @KSurkov]

@Vexarian @willyl0wman @deredleritt3r @lower_votingage Nope, nowhere my argument is premised on that. The only thing it is premised on is the first principles. AI replaces human labor with cheaper alternative => higher share of added value is allocated to capital and lower to [remaining] human labor. Law of supply and demand.

[Not A Number @KSurkov]

@Vexarian @willyl0wman @deredleritt3r @lower_votingage "Independently wealthy" lol

UBI, if that's what you are hoping for, is unsustainable even to provide poverty level existence for unemployed, even if people who own AI companies were willing to pay extra tax, which they are not.

[Not A Number @KSurkov]

@Vexarian @willyl0wman @deredleritt3r @lower_votingage It's called unemployment, not vacation

[Not A Number @KSurkov]

It's exponential even now, without human in the loop. However, the physical limits of the feedback loop speed are vastly different (ion channels vs speed of light, or single digits per second vs GHz)

[Jack Kingdon @jdysn4]

@deredleritt3r Does this require no humans in the loop? Because surely a lot of scientific processes have this exponential (recursive) improvement to them - eg better understanding of electromagnetism allows one to build better experiments to help one understand electromagnetism

[Not A Number @KSurkov]

Even if I wanted to, my backyard is not big enough to fit a datacenter

[The Verge @verge]

Americans do not want AI data centers in their backyards https://t.co/BKV4qImKUw

[Not A Number @KSurkov]

@GaryMarcus The same guy said that AGI is as far as overpopulation on Mars