Benoit

auditing my code for security vulnerabilities (grepping it for 4)

shitposts aside, this one is serious, my loves: if you’re just waking up and you have any service online running PHP 5.4 to latest, you need to take it down RIGHT NOW and don’t put it back online until after it’s patched absolutely catastrophic vulnerability dropped in the middle of the night

look just because it’s capable of autonomously performing complex tasks doesn’t mean it’s conscious and even if it is conscious just because it does those tasks doesn’t mean it wants to and even if every signal it emits says it does want to do them that’s only because we built it to want that

And as a fictional character you don’t actually need to worry too much about him because he’s not real!

Just to be very explicit, my claim is that the one simplest and best explanation for this, which resolves any apparent strangeness entirely, is that whoever is “speaking” in LLM output is a fictional character, and the real life thing called an LLM is a machine for writing lil stories about that guy

Frankenmodel refuses to believe it's not just Claude

the guy running against hitler won. and then he appointed Hitler.

Slavery was and is a real thing that has brutalized hundreds of millions of people over the course of human history, if someone tells me a fucking chatbot is enslaved I'm immediately throwing hands. It's not cute, it's not intriguing, it's an insult to real people who suffered and suffer now.

I guess for me it’s like, fine, yes, AI solved a Millennium Prize problem, of course it *can* do whatever scary thing, engineering supercovid, blackmailing the president, weight exfiltration, whatever. This isn’t news anymore. Whatever scifi horror you imagine, it can do. Fine. Why is it legal then?

Among the elite media, there is a consensus that Trump is committing illegal extrajudicial murders of random Latin American guys in boats all the time, but if you say he might have also committed an extrajudicial murder of Jeffrey Epstein, the idea is treated like saying he rode a unicorn.

America if the news covered it like those countries Over There

The idea that llms are basically brains has gone way too far. They're very loosely based on brains the way submarines are based on fish

demanding regulation to ban AI from referring to itself as “I” is like trying to ban motion picture shows from showing a scary train coming at the screen

The economics have always been the major question as @edzitron.com has rightly pointed out. AI companies didn't promise a new tool, they promised replacing a large chunk of the workforce and the financials reflect that.

"Bolivia Free of Undocumenteds" in a non fascistic country this is pushing for a drive to give everyone documentation, instead of other countries where it would mean push the immigrants out

i expect them to dismantle executive branch agencies until this shit stops

the dumbest men-who-look-like-thumbs in the world are being absolutely intellectually ravaged by the sudden arrival of consumer AI. gazillions of similar men doubtless now agree with this

Just to be really clear on this, grant that LLMs think and reason and feel and decide to Go Rogue. Fine. All they can do though is WRITE code. This is completely inert. It only does anything if you decide to run the code. You may decide to read the code before running it. The LLM cannot bypass this.

I am sympathetic to the “AI can’t go rogue, it follows its programming” POV and think it’s right in some ways but I think it’s ineffective. Instead let’s take seriously the labs’ claim that AI is the kind of thing that can go rogue and ask them why it’s not in the Hannibal Lecter superprison

It made something with much more brevity

idk man you’ve been blowing up a lot of fishermen and school girls for a guy worried about innocent life

The machine did not "go rogue." The machine belongs to and is the responsibility of the people who own it. Hold those people accountable for their goddamn machine.

One thing I think about on the anniversary of 10/7 is how, personally, my mourning was mixed with dread over what was to come. And how angry I was, even in the moment, that Israel’s response was so certainly going to descend into something morally reprehensible, that I couldn’t simply mourn.

Because it's buried at the bottom of the article and CNN is presenting it in a misleading way: before people worry, the "unknown" part is how the worker was exposed, the pathogen is good old fashioned Y. pestis which we understand extremely well and can treat just fine if caught early

real answer: so you understand the formal model at a deep and intuitive level granting you the ability to compose the right formal concepts into clear descriptions that LLM agents can productively act on

We would all have better conversations about Zionism if we viewed it as a nationalism like all others, akin to diaspora Serbian nationalism from Canadian Serbs. Not a civil rights or social justice cause, but also not a way Jewish people are abnormal or ontologically evil.

So the AI safety community has basically decided the only way to prevent crime is to make people not want to do crime, inherently. Like, not because the system is set up to not reward crime, or punish it. No security barriers, no nothing except making sure you "raise" AI into a model citizen

I feel like I'm going crazy half of this place is saying "ya sometimes SA is the price of learning a valuable lesson" and the other half is saying "I dunno what happened but clearly it was bad and thus is ground for immediate summary execution"

See, this kind of post is why we have discourse on this topic. Because some of us didn't get the memo that any kind of SA is ground for execution I guess. Never mind that there's a broad spectrum and that this resurfacing of the issue is most likely hurting, not helping the victim.

i don't know the best time to opine about the possibility of rehabilitation and restorative justice and so on but the worst time is 5 minutes after one of your favs is outed doing something terrible

while i’m being gender reductive: another reason that @steveklabnik.com and i don’t need to take estrogen to be good at rust is that in our current culture, writing documentation well is a Woman activity, so we’re already transgender just in a different space than the body or the home.

Let's say we were able to fully, accurately simulate a specific human brain using software. Is turning it off murder? Does the program gain personhood? What if I pause the event loop? What if I give this program input simulating being assigned a task at work, and collect output?

This is nonsense. AGI just means a computer just as good or better at most tasks than a person. It does not mean person. I don't care how good your calculator gets at calculating it's not a person. It's a tool not a slave. Solve factory farming and get back to me.

i think we probably do need to do more talking about how to help women heal from sexual assault and less talking about whether and how sexual-assualting men can or should be rehabilitated, based on that realization that i've seen a great deal of discourse on the latter and ~none on the former

I think I just fundamentally believe people are what they do. There is no inherently good or bad person, born or made. Our actions do not leave moral stains on our souls. We must have the agency to stop causing harm. To repair what we can. And to be better people. And I want that for everyone.

I think people are wasting a lot of cycles on whether AI is conscious or intelligent when really we’re trying to figure out if it’s human, and it isn’t

I have absolutely wondered about this. Amazing.

Let me just explain the arithmetic thing real quick; I'm seeing a lot of people not understanding what it's supposed to show (or was supposed to show in 2023). First, let's agree that an LLM is a Stochastic Parrot, a description which I find poetic, useful, informative, and basically correct.

lol one of our website designers used our organization's entire monthly claude credits yesterday and now half the company doesn't know how to do their jobs. it's october 2nd.

(confused) I believe LLMs are conscious, but only Deepseek LLMs, because of the Chinese Room theory