Everchange

[Everchange @evereverchange]

@ValsTutor @NunoSempere I'd have expected there to just not be that much at training for writing well specifically. For instance, opus got significantly better at writing maths papers/reasoning, and I think that generalized from their specific math capabilities training

[Joe @JoePostingg]

If you simply froze AI capability at Mythos / Astra levels and focused on decreasing the cost of compute you would totally revolutionize software development with zero existential risk

[Everchange @evereverchange]

@allTheYud Yep, it's an edit of Boston's original video https://t.co/AlF9TloIjR

[EigenGender 🔸 @EigenGender]

you will burn all the things that you think you care about at the altar of the thing that you Actually Care About

[Everchange @evereverchange]

@andres_swp https://t.co/QoKD04nIgQ Seems very relevant here

[Eliezer Yudkowsky @allTheYud]

Science: So there's a guy who made some successful predictions way ahead of other people, and your complaint is --
Msr. Grundy: That they're weird. Why do they need to be WEIRD?
Science: Grundy, we've had this conversation many times previously over the course of human history. People who make unusually correct predictions are, usually, *unusual*.

Grundy: I don't like the WAY this guy is unusual. It's offensive!

Science: Offensive how?

Grundy: Well, they believe all these OTHER weird things! Not just the one thing they made the successful prediction about! And some of the things they believe are offensive!

Science: Msr. Grundy, how were you treating people who were making this prediction 15 years earlier, before it had come true and before everyone else believed it?

Grundy: Like they were weird dangerous people who believed crazy things, which is to say, like animate pieces of shit I didn't want getting on me.

Science: There you go.

Grundy: Go where?

Science: The first person or people to openly make this prediction needed to not be swayed by what you'd think of them, or how you'd be triggered to treat them badly for believing what they did. They needed the skill of ignoring the internal pressure of wanting to believe something that would let them *not* be social outcasts, or maybe to just not feel much of that pressure; so they could figure out what to believe based on the small quiet voice of Reality, that otherwise could've been drowned out by a dozen fine rationalizations to let them believe something less triggering.

Grundy: Why can't they just LIE about everything that would make me unhappy to hear? Like a normal person!

Science: Yes, that is definitely a very normal thing for a person to do! People like you're describing, if they believe something unusual, will keep it to themselves. They won't make a big public thing out of believing it --

Grundy: Right! That's how a decent person should behave.

Science: Which means you don't get to hear about their weird advance scientific predictions.

Grundy: Why can't they only talk in public about science stuff? I think I could manage to ignore it if it was just science technobabble they insisted on being weird about.

Science: Strangely enough, the people who get really invested in a subject, enough to make successful contrarian advance predictions about it, generally don't feel that the thing they're interested in is mere science technobabble. It feels as important to them as your socially mandatory beliefs feel to you. Often the wronger people in that field of science or human endeavor don't think it's unimportant either, and will make them outcasts for their weird prediction.

Grundy: So what you're saying is that they're just incompetent at looking normal?

Science: Or choose not to. But yes, often it's both. Very often.

Grundy: Why would be bad at looking normal often go together with a choice to not look normal? I don't get it.

Science: And that's even before taking into account the more famous cases, where the revolutionary idea was important to the public; where you're not conditioning on making an unusual correct contrarian prediction about an obscure matter that only scientists would understand, but taking an unusual stance on a big-deal issue that normal people and Grundies could have 'yikes you must be insane' reactions to 15 years earlier.

Grundy: I don't understand why they have to shove all these *other* weird beliefs in my face.

Science: Because the behavior of shutting up about controversial things you believe in and keeping them to yourself, has historically anticorrelated with being determined enough to study them for the many years required to git gud. You are describing a coherent personality, but it is not the usual personality of people who went on voicing an unusual true prediction even after it triggered Grundies to be mean to them.

Grundy: I'm not okay with this state of the world. I'm not okay with how this person is getting credit for an unusual advance prediction, which sounds like something that only a sage serious person should be able to do, like a science version of a wise old shaman, and instead they're acting like this giant weirdo.

Science: Everyone who knew how to act to you like a very sage serious person, and who chose to act like that, predicted to you the things that very sage serious people were supposed to predict. If they talked about AI they probably said that AI was going to happen in 2050, far enough away to not seem shocking to you. Looking like you believe a high-status person should look is a capability, a skill, and a choice. Sometimes in scientific history there have been revolutionaries who defied what everyone else believed, and were right, who *also* had the capability and developed the skill and made the choice to otherwise comport themselves as very serious respectable people who believed all the other things Grundies want them to believe. It's sometimes happened! But it has not been the normal case in science.

Grundy: Are you sure it's not normal?

Science: Msr. Grundy, you really need to study more science history. This has happened before. This has happened before a lot.

[Everchange @evereverchange]

@naturalSmartnes @haider1 Fable 5.5 has been on stealth release for a while now.

[Everchange @evereverchange]

@danallison Seems like you're talking about the whole of Functional Decision Theory? My favorite simple example of pseudo-retrocausality is https://t.co/mNxvZmuq94

[Everchange @evereverchange]

@sergeantsup I've found very often that the skill of holding cognitive dissonance, of being able to hold the possibility that what you're doing is monstrous "and that's okay"/you-don't-need-to-change-it-at-this-moment to be very important to be able to change your mind and actual behavior.

[Everchange @evereverchange]

@PosaniRishi @Samaytwt That's because that's the case

[KingTalon23🇨🇦 @KingTalon23]

@AdrianDittmann

[Everchange @evereverchange]

@burny_tech From LLMs, we've seen a lot of combinatorial, and we're starting to see a bit of exploratory. I think the first transformational example will be quite a shock.

[Everchange @evereverchange]

@anord_ink Well the replies do confirm that I'm very lightly colorblind, so thanks for the insight! (I cannot see any of them either)

[Everchange @evereverchange]

@LVTYIMBY I'm confused is LVT actually unpopular to most libertarians? Everyone I've spoken to seem quite in favor of it, though some think it's not enough

[Everchange @evereverchange]

@burny_tech (tbc I expect this to scale as we continue giving more parameters, but that's the shape that I still see Opus 5.5 and Astra fumbling about)

[Everchange @evereverchange]

@burny_tech It's like they have a very hard time of seeing the thing without a specific question to resolve, and the question they consider are often too low level.

[treetown @treetowntree]

@elon_lit

[Everchange @evereverchange]

@itsmeMehak_ Imagine I give you a 1000 line file, and ask you to copy it word for word, except for changing a single number. Doesn't sound very appealing. (tbf, this is mostly a problem of https://t.co/yGVazGXLsh not having native linux tools. Claude code would accept because there's sed)