Finally figured out how to talk to people about AI danger:

We know it works because this reel went somewhat viral on insta.

(@bryan_johnson this is inspired by you, fyi. Let's not die.)

Very good! ✅

if you really cared about convincing the public you wouldn't post something sneery pandering to your ingroup

if anyone builds it, everyone goes night-night

hidden somewhere in lighthaven, there's a children's book titled "Goodnight Doom"

This comes off as extremely condescending, and I anticipate more people would come off as not trusting you than convinced of your argument from this.

the condescension is part of a good-spirited joke that most viewers will recognize; see e.g. for context. it's a tiktok trend these days.

I don’t have as visceral a reaction to Bryans delivery as yours. Idk if it’s subject matter, familiarity, or intonation. Best guess is you’re asking me to follow a logical argument, and Bryan is just listing instruction. I think it causes some dissonance. Would be curious if others feel the same.

I know this is a second-derivative joke, but it deems people not listening as "dumb", which I think is an incorrect message

it's part of a tiktok trend (started by health folks, I think?) which takes some of the edge off. And also it channels some real annoyance, which I think is actually healthy (relative to hiding it).

Believe me I share the annoyance (see my feed), but the "our political adversaries are dumb" underplays the problem, they're typically very smart but it's the motivated complex reasoning that 's the issue, and it's worth flagging as such, I think TT would be good for that!

I don't think the implication is "our political adversaries are dumb". I think it's somewhere between "I know you wanted this info so I'll give it to you in a super simple way, but with side of humor" and "aren't you frustrated that everyone around you misses the obvious?"

I'll take your word for it, but that's not how it read to me: - uninformed public does not need small sentences, they deserve (and will be more receptive to) more respect - motivated reasoners will not be receptive to this either - did I miss a class of ppl you're talking to?

the uninformed public is smart enough to get the joke, and does not think you're condescending to them (as opposed to their surrounding environment, which most agree is insane rn). Especially in the context of the larger trend.

Maybe this is my misreading: I see the joke as aimed at motivated reasoners, which is an inside joke not for uninformed public. Did I get that wrong?

i think so! at least if by "motivated reasoners" you mean "people at the labs" this might be helpful context: the trend isn't usually aimed at analogs of lab employees, and much of the general public audience that sees it will be familiar w/ the trend.

I see you are preparing to talk to Congress and the Senate😁

some congressional offices are actually really on-the-ball these days! some.

Idk, bro, I am a politics Junkie, and I am still to hear a single politician besides Bernie take AI with the seriousness and urgency it demands. I really hope you guys actually make it in front of them sooner rather than later for obvious reasons 🤒😅

@allTheYud why use many word when few word do trick

…

I don't think the problem is the complexity of the core argument. It's bigotry from the masses. If the public aren't cultivated to accept a domain of information they're not familiar with they won't. As an example, if quantum physics (wormholes, entanglement, super-position) wasn't culturally normalized then most people would treat anyone trying to profess it like they were a schizophrenic. Schizophrenics have broken brains and present uninterpretable "information". Bigoted normies don't acclimatise themselves into unfamiliar domains, they just categorize it as 'schizo-babble' or 'expert approved soyence 2 hard 4 me' and disengage. 95% of the public, including people with significantly above average IQ"s are somewhat- profoundly bigoted in this way.

eh, i think the masses can follow "if big corporations make superintelligent machines that'll probably go wrong for us somehow". it's common sense. i think the issue is more that everyone thinks everyone else has forgotten that common sense and buys an "inevitability" narrative

it's not actually inevitable that we build this stuff. that's just what the big players want you to think.

Of the things you’ve written and made, this is my favorite! More content like this please

.@plzdontkillus looks like you're gonna have to run another one

did eliezer write 300% of this script?

> We stop AI > We die > Very very bad

Simply stopping is not a solution, the opportunity cost of not benefiting from an aligned ASI is immeasurable

defeatest mindset. world leaders haven't realized that AI companies are actually, really, risking the planet. they might put a stop to it real quick when they notice, like how the trump admin slapped export controls on fable with like 90min notice. don't give up before you try.

What are you talking about? I never said it wasn't possible to stop, I said it wasn't a solution, we can't just ignore the incredible benefits of a safe ASI, you're the defeatist here thinking we should just stop instead of solving this alignment problem

@So8res Nate thinks we should solve the alignment problem and build ASI (as do I). He just thinks we're not in a position to do so today (especially under race conditions), so we need to internationally halt development for a long time in order to achieve that goal.

@DIMENSION_YT yep. sorry for misinterpreting you as saying "stopping isn't possible"; i rounded you off to something i hear a lot. i agree we can't ignore the benefits; but we don't get the benefits by committing planetary suicide; if we want them we've gotta stop & find a different route.

I have a hard time seeing how this problem could be solved through theory alone, it seems more realistic to me to move forward at a reasonable pace, analyzing what's happening and the effectiveness of the different solutions as they gradually become more capable, is that what you mean by "stopping"?

@robbensinger i think that if we race ahead growing smarter and smarter versions of black-box AIs that regularly act in ways they know we didn't intend, we just die. there's other ways to build smart machines. also, there's ways to create smarter humans. rushing is lethal; let's take our time.

West stop ai advancement. Bad. China keeps going. Machines don’t take over and they have stronger technology. You work in bauxite mine. Bad.

love it

@ifonlyalabama

lmao

This is so unbelievably condescending and will make normal people justifiably hate you.

it’s hard to pull off humor on this topic since there’s so much at stake, great job landing it some are calling this condescending, but i find your delivery likable. even if you were my opponent in this debate i’d see this and go “ok that was great actually”

it’s giving this

And somehow people will manage to misunderstand

it's easier when they're getting paid to

i read you and yud's book and didnt really understand it i understand this me no want unaligned machine consume light cone

Your goals != my goals. Should we rally to "stop each other" or think how to ensure our goals do not conflict? Also, LLMs right now do not have innate goals - humans are guiding RL (directly or through another model that internalized human preferences) and prompt at inference

You are a genius and my favorite blogger from this day on please share all the social links so that humans can subscribe 🤩

i'm @so8res in most places but i'm nateasoares on insta (because i got the so8res handle there Too Early and it somehow wound up broken in a way where any attempt to interact with it just causes errors)

❤️❤️❤️

Video very good big success 🙌 ✅️

@SimonLermenAI Ok this is pretty funny

This is Brainrot, that’s why it went viral.

Stopping very hard. Stopping hurt self. Good people stop make bad people go faster. Stopping plan bad. Find another way.

What if big AI good, and stop Ai bad?

maybe a whole series like this: - why not make ai same goals as human? - why think ai can get big big smart? - why not put ai in strong box for not escape? etc

communists/psychopaths believe others having more (power) means they have less, in reality the more you help others the more you help yourself.

This is a good presentation of the argument imo - the forced simplicity makes it clearer. “AI learn goals” is the weakest part imo.

this will give people a false perception of what is happening instead of actually educating them. I wrote about this too and how I see stuff like this growing to a greater scale in the future.

Except for the part where it is literally impossible to stop, this is the thing alright. 🤷‍♂️