Ben Hilty

Ben Hilty

@benhilty.bsky.social

Fuck this.

418 followers600 following225 posts

To me, all the people freaking out about AI becoming conscious are about the same as someone staring at a magic eye image of a horse until the horse pops out at them, and then going into hysterics about how there's a horse trapped in there and we need to help it.

Ben Hilty reposted

Sat down to write a YouTube comment, and ended up writing a short essay. So here's my short essay about how "AI" chatbots work and "AI" consciousness:

Ben Hilty@benhilty.bsky.social

I can't emphasize enough how these things are not conscious.

Sat down to write a YouTube comment, and ended up writing a short essay. So here's my short essay about how "AI" chatbots work and "AI" consciousness:

Ben Hilty@benhilty.bsky.social

I can't emphasize enough how these things are not conscious.

View full thread

P.S. I wrote this by hand, but I tried pasting it into Gemini and asking if it was accurate, and it told me it was "spot on" and "highly accurate." LOL GET REKT

View full thread

It isn't magic. It isn't any kind of intelligence. It's just a crapload of statistics. It's kind of amazing what it's been capable of from that one trick. (Granted, that one trick is insanely complicated.) It might almost be something cool except that it's all built on stolen labor, its impact on

humanity is devastating because of its misuse and the misunderstanding of its nature, and that a drug-addled billionaire death cult basically worships it. (Literally. Look up Roko's Basilisk. This is the actual reason some of them want to build it. I'm not kidding.)

View full thread

the prompt for every request to throw the statistics off. That's also why it hilariously caused it to bring those things up seemingly "unprompted" when it was actually just being added to the prompt for every request behind the scenes. Adding it to the prompt on the server puts a

View full thread

intention. Every token in the input text tugs at the probability for what the next output token will be, based on an absolute sh*tload of numbers from the training process. The numbers define the statistical relationships between words and phrases based on the stolen training

data. That's why Elon can't make a version of Grok that isn't just naturally a N*zi, because statistically most of the text that was fed into it for training wasn't N*zi sh*t. It only morphs into MechaHitler and starts spouting off about white genocide when he explicitly includes that in

View full thread

the document would be the task it's working on. Just running the same function over and over again in a loop having it produce more text, but now some of the text could be a command to execute somewhere or some code to compile and run, all produced by statistical probability.

It doesn't know what the end of a sentence will be when it starts the sentence because it's literally producing a part of a word at a time, going from left to right, adding to what's there until it's made a sentence (or a line of code, etc.). Repeat. No thought goes into it. There's no

View full thread

It's literally the only thing it does. Everything else is just a fancy way of using this one function that it has, by creatively manipulating the input and output.

It's just appending tokens over and over to create a new paragraph or a new summary or whatever output the prompt asked it for, and then doing that over and over again with each new prompt to add to the document that is your "conversation." Or in the case of an "agentic" AI,

View full thread

Like I said, there's nothing there to remember anything, nothing there to have thoughts. There isn't even a "there!" "There" could be in a different state or even country from one message to the next depending how the servers are arranged. No matter which server gets the prompt, it

takes your conversation (or summary of it, etc.) along with your prompt, feeds that in as the input, and then generates a new message to add to the conversation. It predicts the next token over and over again until it's finished producing its output.

View full thread

other message could be coming from a different server. The only reason it appears to be a coherent conversation is because it's loading the stored conversation as part of the input and reading that as the "context" for the conversation. (I say loading "the conversation," but it may

be a summary of it, or otherwise compressed version of it for keeping the text at a manageable size, and which is why things start to fall out of the context over the course of longer sessions.)

View full thread

A request is received. A (any!) server takes the text from that request, does a bunch of calculations on it (like literally a hundred trillion!), and responds with some new text.

It's very likely when you're "talking" to "one" of these things that you aren't even hitting the same server from message to message. Servers are set up in clusters and the request load is balanced between many servers to be able to handle large numbers of requests at once. Every

There's nothing there to be conscious. They're just having it read in everything in the conversation that's been said as a document, and asking it to add to the end of the document. It takes in all that text, transforms it based on the statistical weights from training on all the stolen data,

and produces the next chunk of text one token (basically parts of words) at a time. There's nothing there, nothing happening when it's not translating the input text to the output text. No thoughts. No memory.

Ben Hilty reposted

the big AI danger is not companies suddenly inventing skynet - that's marketing rubbish, musk promising mars colonies by 2030 type stuff. the danger is companies claiming to have invented skynet and then putting what they actually have, a chatbot, in charge of stuff that's important

Ben Hilty reposted

Ben Hilty reposted

Today in @thenation.com: a harrowing piece written from inside the Dilley ICE detention center by Maryam Tahmasebi, a green card holder who's been locked up along with her husband and son for months despite none of them committing any crime. www.thenation.com/article/soci...

thenation.comOne Woman’s Message From the Hell of ICE DetentionMy husband, son, and I have been locked up for months—despite being permanent residents with no criminal record. It is an unendurable, unending nightmare.

Ben Hilty reposted

Michigan Governor caught on hot mic with Oracle CEO Clay Magouyrk, discussing citizen opposition to AI data centers, bragging how she ignores their concerns: “We’re used to people saying 'f*ck no,' & doing it anyway.”🇺🇸

Congratulations, Mr. President, on single-handedly reducing global fossil fuel use by double-digit percentages and forcing increased adoption of electrification and renewables. And you did it all without creating any new jobs or economic benefits to Americans. It's the Green No Deal.

I fear that things are going to get much worse than people are expecting much faster than people are expecting and for much longer than people are expecting.

Ben Hilty reposted

Minneapolis celebrates spring with a parade of giant handmade socialist puppets and a festival where people row the sun across the lake. This year's Mayday festival was today, and I'm going to try to share some photos and videos (many taken by others) to try to explain it for my non-local friends.

Ironically, Russia, China, and Iran are some of the countries best positioned not to be—as they are so fond of saying—sent back to the Stone Age by our actions.

Big Snake Oil

Powerful anti-vaccine advocates and people selling potentially harmful goods such as raw milk are benefitting – financially and otherwise – from the push to write anti-science policies into law across the U.S.

Ben Hilty reposted

I am quoted here so let me be totally clear on my take: Jonathan Greenblatt has made a calculated choice to aid the consolidation of authoritarian power by the MAGA movement in exchange for the suppression of criticism of Israel. And the consequences for American Jews will be devastating.

Noah Shachtman@noahshachtman.bsky.social

I spoke with 40+ sources for a look inside the Anti-Defamation League. Here’s some of what I found. ADL’s gentle response to Elon Musk’s infamous salute was partially driven by his support of Israel. To ADL, that indicated he didn’t have “antisemitic tendencies.” (1/x) nymag.com/intelligence...