maplethorpe

It's not intelligent.

I could theoretically do the same math on a piece of paper to generate a list of tokens, then decode all the tokens into a sentence and read the sentence I generated for the first time at the very end of the process.

In this process, where was the intelligence? It didn't come from me. I didn't know what sentence I was generating until I was able to read it at the end. Was it in the pencil? The paper?

> And they sure are effective—it’s very difficult to walk by a billboard that appears to be popping out of the wall without being captivated by it.

This isn't really how they work, in my experience. In reality, people on the street who are interested in seeing the billboard attempt to gather at a single vantage point where the illusion works. If you stand anywhere else on the sidewalk, the image becomes distorted, and the illusion breaks.

My guess is that the trucks in question exploit the fact that when I'm driving behind one of them, I'm stuck at that single vantage point where the illusion works.

I used to be proud to say I worked in tech. Now I look around me and I'm surrounded by scam artists and grifters. Where did it all go wrong?

I can't tell if you're being facetious, but a future AI really may be able to fill in a missing database. Like, if it knew some of the entries, it could infer the rest.

Mythos apparently wrote a poem so beautiful it made Dario cry.

A user is someone who uses your product, not someone who wants your product.

Think about it on a micro level. Do you work better when your coworkers are thoughtful, or when they're thoughtless?

Now multiply that by a billion, and that's why it's good for the economy.

Don't have a billion users if you can't offer them support?

I didn't say you said all technology was good. I'm addressing the part of your argument where you point to past events as evidence that the automation of labor will always continue to be good in the future. I used an extreme example to show that one day, if a new technology that automates labor is actually a net negative for society, your philosophy won't catch it. It will slip right through.

To spell it out so we don't keep going around in this circle: it's worth looking at indicators other than the past to judge a technology's benefit to the labor force.

This + NFT integration will be the real game changer. Like it's Breaking Bad, except Walter White is decked out like one of your Slonks. Or it's Indiana Jones stealing a Bored Ape instead of the idol. Possibilities are endless.

I'm not trying to argue whether AI is good or bad. I'm only pointing out the flaw in your philosophy.

Like, there will one day be a technology invented that could indeed wipe us all out in 5 years. On a long enough timeline, it's a certainty that someone will come up with such a thing. And when it comes, there will be people such as yourself saying "no technology has ever wiped us out before, therefore this one won't either". And then it will wipe us out, and there will be no one left to say "well I guess this time was the exception".

Theoretically, if a technology came along that destroyed the human condition in a four year time frame, your "let's wait longer than four years and see" philosophy would kill us all.

Mine was a joke too, sorry. I was just trying to go along with your bit.

I don't know if you're joking but FartCoin hasn't shown returns all year.

It's the same reason young men are drawn to crypto. Younger generations are faced with an economy that prices them out of the housing market, so they feel the need to explore alternative wealth-building pathways if they're to achieve the aspirational lifestyles they've been sold.

They should try releasing one with a futuristic space-age design and flashing rainbow lights. Maybe give it a name like HARDTEK and put some random techy shapes all over it.

> And this line to me is meaningless unless we have specific definitions of what the Christ Wedge meant by "animation" and "just effects"

I'm not really sure what you're trying to say, sorry. What else do you think he might have meant?

Thanks so much for this! I still haven't got around to building my own language model yet, so I'm a bit fuzzy on the details, but if I imagined a thought experiment where I did all the math by hand on paper, I just couldn't see how I would end up with a different output each time given the same inputs. Finding out that the variance other people are seeing comes from the server/hardware stuff clears that up.

This is a surprisingly annoying question to Google. A lot of articles give the reason that softmax returns a probability distribution, as if the presence of the word "probability" means the tokens will be different every time.

Oh.

Can someone explain why this is? Do LLMs somehow contain a true random number generator? Why wouldn't they produce the same outputs given the same inputs (even temperature)?

edit: I'm not talking about an LLM as accessed through a provider. I'm just talking about using a model directly. Why wouldn't that be deterministic?

Abstraction has a specific meaning in computer programming. I don't think he's misrepresenting it.

https://en.wikipedia.org/wiki/Abstraction_(computer_science)

I found some more details about this, for anyone interested. It looks like critics of Tron's visual effects mistakenly thought the computer was generating it all for them, with little human input, when it was actually quite a laborious process.

"Tron’s offices were trailers in the Disney parking lot, recalls Chris Wedge, then an animator for MAGI, who worked on Tron’s light cycle sequences. “[That’s] because the Disney animation department didn’t believe that this was animation,” he says. “They thought it was computers just making effects. They just didn’t understand anything about it.”"

"Tron’s distinctive glowing circuitry was achieved through a technique called backlight animation, which involves making a negative of each frame and hand-painting the glowing areas. There were 75,000 frames to do; more than half a million pieces of artwork."

"Star Wars and Alien both feature 3D wireframe graphics projected on screens. Only a few companies could produce such images, each of which had their own room-sized computer and their own custom-built software. The process was still cumbersome. “We had to figure out how to position and render objects 24 times to make one second of perceived movement on the screen,” says Bill Kroyer, Tron’s head of computer animation. Tron’s animators had to map out the CGI scenes on graph paper, then calculate the coordinates and angles for each element in each frame."

https://www.theguardian.com/film/2022/jul/05/tron-steven-lis...

I read through some of the sources in your link, but they don't paint as clear a picture as you claim. Yes, the cost of inference appears to be coming down, but we so far don't really know why that is and what the largest contributing factors are. With other costs rising (e.g. the rising cost of training, the cost of inference scaling with number of parameters, and reasoning models using more and more tokens), it means we can't yet make any certain claims about long-term economic viability. There just isn't enough data yet.

Taking a look at the sources in your link, the MIRI's "Observations About LLM Inference Pricing"report [0] seems like one of the least biased ones (forgive me if I don't believe everything a16z has to say about the economics of AI).

Some choice quotes from the report:

"Imagine you went to the gas station and the price was $4.00, and you look across the street at another gas station and the price is $40.00 — that’s basically the situation we currently see with LLM inference."

"Overall, LLMs do not appear to be priced like other commodities."

"It's possible that some providers are slightly modifying a model that they are serving for inference, for example by quantizing some of the computation"

"Unfortunately, it is difficult to make strong conclusions about the underlying costs of LLM inference because prices range substantially across providers. The data used in this analysis is narrow, so I recommend against coming to strong conclusions solely on its basis."

Another source you linked, Don't Panic Labs[1], seems to agree with Zitron:

"It is a little unclear as to why the price per token is dropping, and I am still a little worried that the price per token will, at some point, go up."

"According to another graph at Epoch AI, the cost to train a model doubles every eight months. This tends to align with the common wisdom that we are getting a really good deal right now, while everyone is fighting to build market share."

If your inference costs come down due to quantization, that doesn't count, since you're cutting costs by offering a worse service, and there's only so much you can do that before your customers walk away. If your inference costs come down due to subsidization, that doesn't count either, since that obviously won't last forever. If your inference costs come down but your training costs double every eight months, that poses a significant problem for your business. If your argument to that is "training costs won't continue to increase at this rate forever". Well, inference costs won't continue to come down at this rate forever, either.

From what I can tell, there still isn't enough data to draw a strong conclusion either way.

[0]: https://techgov.intelligence.org/blog/observations-about-llm...

[1]: https://dontpaniclabs.com/blog/post/2025/12/02/the-price-per...

HN is an interesting case, because a lot of those types of comments are likely to be either astroturfing, people who work at those companies, or people who are invested in them. I wouldn't extrapolate too much to the rest of humanity based on the people here.

> If nothing else, to have a shared experience with other people. A lot of the value for people is derived from the fact that they can talk about the same song with someone else. If it's all individualized you lose that.

While I agree this is a valuable part of the human experience, I don't think the average person would recognise it as one. Without it, they might start to feel a general lack of connection to the people around them, but probably wouldn't be able to trace where it came from.

It makes me wonder what other human cultural experiences we've lost over the years, which are causing us a kind of collective mental anguish in their absence.

In the Uber Eats app I can't even request a refund for an incorrect order anymore, because the UI doesn't allow me to scroll down to the "submit" button.

It's been like this for months. I finally got my explanation.

Why isn't Claude coding that stuff too? Is Claude only good at coding Claude?

What are all their SWEs doing, if Claude is coding itself? And why are there hundreds of open SWE positions on their careers page?

I don't think I'm really following your analogy. Did I put it out on the side of the road because I didn't want it anymore, or did that person steal it off my property?

From what I can tell, the million dollars does not belong to me, but the person who stole my statue should be held accountable for stealing it, and they should be held accountable for selling an object that did not belong to them. Both of those things should be illegal.

Your analogy seems besides the point, though. In my original question, I specifically mentioned that the art piece was made for fun. Sometimes, as an artist, I make things just for fun, with no intention to ever sell them. Other people should not be able to take the things I make and sell them without my permission, or without some sort of deal beind struck between me and the seller. A society where it's legal to take another artist's work and sell it without their permission would be supremely unfair. A society for vultures.

What about something you've made for fun but haven't made any money from? Should someone else be allowed to sell and profit from your work?

I'm not expecting to be "passively paid" for my hobbies. But I'm expecting that someone won't steal and profit from the things I make. Why would that be fair?