Colin: Wokester

United States Secretary of Transportation Sean Duffy is seen wearing a blue suit, white shirt, and tan shoes. He's shaking hands with someone in a black outfit.

A close up of his shoes show weird stitching on the side. They look like they're the first shoes ever made in history.

can't believe this ancient mesopotamia shoe wearing ass said people need to dress up at the airport

I continue to advocate, at a minimum, for nationalizing WalMart w/o compensating the Waltons (they've made enough).

The choke points on critical infrastructure being privatized are exclusively serving roughly 1,000 people. The price of that greed is passed on to the other 339,999,000 of us.

rather than another PSA, try this out:

18 U.S.C. § 115 - THREATENING PUBLIC OFFICIALS

Bill Cassidy should be required to have this shit framed and hung where he can see it every time he sits at his desk

this is the bargain you made, asshole

you knew better and you did this to the rest of us anyway

In recent days, Trump has repeatedly posted his desire to have a state AI law moratorium, and reiterated it on Wednesday during his appearance at the US-Saudi Investment Forum, couching it as a way to fight “woke” ideology. “You can’t go through 50 states. You have to get one approval. Fifty is a disaster. Because you’ll have one woke state and you’ll have to do all woke. You’ll be back in the woke business. We don’t have woke anymore in this country. It’s virtually illegal. You’ll have a couple of wokesters.” As part of the AI Action Plan released earlier this year, Trump had directed several federal agencies, including the FCC, to explore ways that they could circumvent “onerous” state and local regulations in order to promote the industry’s growth and innovation.

This is a direct quote. 🫠

Polymarket tweets: "BREAKING: Zohran Mamdani to require all New York elementary school students to learn Arabic numerals." Laura Loomer replies: "You know things are bad when you can't tell if this is real or a joke. Dear God. It's real. Isn't it?"

when I was young, i used to wonder how people like this would fare in life, as they didn't seem suited for gainful employment. it turns out, the internet allowed them to accumulate a massive audience of similar idiots, enriching them and turning them into a presidential advisor

For the record, I'm also not convinced that the current methods for MI are actually useful for any of the purposes you describe, although I will freely admit that I'm not as well versed on this subtopic as I am on others.

I have a degree in computer science and several decades of experience in the field, and I still think you're splitting hairs. If you want to call something that's amenable to MI is a "grey box," that's fine. But I was using the term "black box" in the colloquial sense of the word, and I stand by it.

Ah yes, it is true the word generator is also a copyright violation machine...

And that is the question that the OP was asking. So, as you said, it's incomprehensible, hence it's a black box.

2) Even open models are a black box in that if all you have are the weights (the case for the vast majority of open models), it's impossible to answer the question "what particular document or documents in the training data, combined with prompt P, caused the model to output this particular text?"

This is splitting hairs (and I know how transformers work).

But if you want to split hairs:

1) No, ChatGPT is truly a black box, because the model is closed. So WE will never know anything about how it works.

"Hopefully the word simulator will continue becoming less and less efficient, using more and more power for each request, and then it will finally be as accurate as a simple web search or Wikipedia article."

This, exactly. @frankjeweetwelwie.bsky.social, the fact that you said "the first version hadn't read the book completely" shows that you have no idea how LLMs work. They are very, very frequently wrong, sometimes dangerously so, and at a rate much higher than you think.

Maybe! But we'll never know because the algorithm is a black box. And it's still giving you incorrect information.

The underlying reason is wrong! See my other reply.

In almost every conversation I come across where an AI user is like "look! My AI tool of choice is correct about this thing!" I can find a mistake in the response that the user thinks is correct.

In this case, it's wrong about something that's inconsequential. But the model itself has no concept of "consequentially wrong" or "inconsequentially wrong." It's just wrong, at random, in unpredictable ways.

This is a great example of exactly what I was talking about! ChatGPT is still wrong, in that the year of Susan Calvin's birth is given in the third sentence of the story itself, not in any "internal timeline notes that Asimov provided."

I have at least 10 electrons I am willing to part with if you will also value my company at a trillion dollars.

FWIW, the AP/NORC poll is there.

I hadn’t heard of Gettier before! Thank you!

Gell-Mann amnesia effect - Wikipedia

You will often notice that it gets things wrong, confidently. Sometimes the mistakes are small and relatively inconsequential, but they are frequently there. Now imagine how often this happens for topics you’re not familiar with. This is a form of Gell-Mann amnesia: en.wikipedia.org/wiki/Gell-Ma...

And I’d bet money that even when you think an LLM is giving you a good response, it’s subtly wrong in ways you don’t realize, just because you don’t have a deep knowledge of that topic. If there’s any topic you have deep, decades-long experience with, try asking an LLM about it.

I like to think about it this way: everything an LLM generates is a “hallucination.” It’s a probabilistic word generator. If it’s factually correct in a given response, it is correct only by accident, because by its very nature it does not have an inherent concept of correct or incorrect.

Elon Musk belongs in a prison for the remainder of his pathetic life, and his wealth including the future payouts from this "$1 trillion pay package" should be part of a large public fund established to provide restitution to people adversely effected by the current regime's illegal actions.

this is a far more explicit death threat from people with the actual means to carry it out

I think that advertising (promoting goods or services in exchange for money) should be illegal. Yes, including physical ads like those in magazines and on billboards. Yes, including whatever possible exception you're thinking of right now.

I think people should be allowed to get abortions whenever they want for any reason. I don't give a shit and it doesn't affect me