If someone uses LLM to help wording their idea in a way that's easier to understand, I am fine with it as long as they don't copy-paste it verbatim into their blogpost.
I am getting a little ranty here, but the usual LLM-isms like "not this, but that" or "that's the real footnote here" simply add zero value to the text, and I can't imagine anyone who has proofread the LLM-generated text even once will opt to keep those sloppy paragraphs. If I see them, I can't help but to think that the author had oneshotted the text. And it strongly suggests to me that the author has put zero effort into their work.
I disagree, the problem is the tool. Generating text instead of thinking about what text to use explicitly is not writing, and it's not worthy of someone else reading it.
Fundamental to exchanging ideas is discovering which ideas should be exchanged and how to express them. The process of writing itself is that exercise. If you use a tool to outsource the thinking then you're not thinking.
I definitely think that "not thinking about what you're writing" is a big danger with AI-generated text, but it's not a given. If you generate a block of text and then go through and edit it to better reflect your ideas and voice - that's the thinking process in action. It's also entirely possible to generate slop by manually typing every word - see any corporate communication ever.
I just fundamentally do not agree with you. If you are editing AI generated text to "better reflect your ideas and voice" you are still not using your ideas or your voice. There's no 'thinking process in action' if you've delegated the hardest part of the task to a text predictor!
It really bothers me that people think it's ok to use AI as the starting point for writing. In my mind, it's the worst thing you can do. It poisons the entire effort, and frankly it's disrespectful to give it to anyone to read. Stop doing it!
There are just times that we stumble upon writer's block.
Personally I have had multiple instances of me knowing the idea I want to express, but utterly failing to articulate it in a way that's understood by most people.
And I unfortunately do not have a human assistant, so for a second opinion on how to articulate the idea better, I turn to an LLM. (Yes, I am using my own idea!) I write something, ask the LLM whether it understood what I said, and make adjustments based on that. In this case, the text is 100% written by me, and the LLM is no more different than a random passer-by giving me advices.
But there are also times when the idea is so complicated that I don't really have a good idea on how to untangle all of that. Ideas that live in your brain as a bunch of abstract concepts, if you can get it. If I were to untangle all of that myself - it's fairly good chance that I'll just leave it for "another day" (read: abandoned). I won't be investing any more thinking into it, LLM or not.
So LLM it is. I dump my ideas (as a bunch of incomprehensible human gibberish ;P) into it, make it make sense of the ideas, generate an ideas-untangled paragraph, which I use as an inspiration (or basis, or template, or whichever word you think suits it) for writing my own text with my own voice and my own style. I was able to continue thinking because the LLM keeps my attention from drifting away!
(I always suspected I have some kinda ADHD, but shrug)
I understand where people are coming from when they reject all forms of LLM assistance in writing, it's just that I hope people don't turn it into a dogma of some sort.
This is generally indicative of not having ideas that are worth reading by others. If you can't formulate the idea then you don't have to write about it, because you don't have an idea.
At the end of the day, if your writing is meant to be read by others then using an LLM is a violation of the social contract between you and your audience.
You will NEVER be a good to great writer if you use AI to seed your writing and then edit it.
_Everything_ AI generates, is generalized slop. All of your ideas will be boring and derivative. Mediocre writers will be improved, but the best would only have their ability to think dulled.
Just think through it from the perspective that AI is generating art for you instead of writing. What would all art look like if good artists used it for ideas and "edited" it? It would look like generic, derivative shit. A perpetual crutch that the person can never abandon.
Genuine experiment: Do characters "have a color" for you, maybe? [1][2]
1) "The quick brown fox jumps over the lazy dog." <- Typed by hand.
2) "The quick brown fox jumps over the lazy dog." <- Generated by ai, cut-and-pasted in. [3]
For you, would you classify the former as genuine and the latter as slop?
You'd think I'm being silly, but both yes and no are not unthinkable, and honestly I've actually seen odder arguments in court filings and language standards.
>So if you expect a model that will start correctly telling you info that its local government didn’t want included, this changes nothing.
From experience, the models often do have the knowledge of those topics (strictly talking about the political ones). IMO the refusal is likely to be a product of post-training, as evidenced by various people gaming the prompts just enough to get a proper response out of the vanilla models.
Probably only when you get to things like illicit drugs or NSFL topics, that things will go haywire with the refusals removed.
We really ought to deprecate UTF-16 someday. The fact that it pretends to be a fixed-length encoding has caused all sorts of bugs over the years, with many people assuming n(UTF-16 codepoints) == n(characters) which breaks when the string contains non-BMP characters.
And also, for personal aesthetic reasons I hate that it limits the Unicode codepoint range to an awkward non-power-of-two number (now there are 0x110000 codepoints in total). UTF-8 and UTF-32's 2^31 feels much more natural.
I wonder what it would take. For things like Java, JSA, and c# there would have to be things like a parallel utf8 api (yes please), but the really hard parts is the things that are as old as time (windows).
I don't think it will ever happen, but one can dream.
I doubt most of us would like to complain about anything, but with so many people (ab)using AI for writing their blogposts, it feels that a lot of the Web has become _homogenous_ in terms of style, and it's unfortunate.
>The circle jerking of Chinese models on this site never ceases to amuse me.
As opposed to "ask the model about Tiananmen" which seems to be the site's favorite pastime about Chinese models. ;)
--
Sarcasm aside, I don't think people are happy about _Chinese_ models making advances. They are happy about _open_ models making advances. It's just coincidental that China is the one making them.
If some American lab were to develop a SOTA open model most people here will be equally excited. Although besides GPT-OSS-120B the American labs have been disappointing in this regard.
I think we all know how poor China was in the naughties. They needed foreign investments to enter their country. So now let's think about it: with strict regulations, would have those investors chosen China?
Or do you believe that the Chinese (the people not the government) deserve to live a poor life forever?
China was the 3rd largest economy in the world. Their middle class was fine. If they wanted to add some environmental regulation it wouldnt have halted growth.
Per captia data is not useful because they have so many people who live rural lives without tech. Their urban class was doing well and causing a ton of pollution just like the US does. They then scaled up even more in the least emission efficient ways complete disregard for the damage it would do.
Your partial quote is quite misleading. The article obviously talks about "efficient frontier", not "intelligent frontier".
>In the AI industry, we borrowed the term “efficient frontier” from economists. We use it to talk about managing tradeoffs, most often the tradeoff between cost and capabilities for models. A model is a “frontier model” if it offers the highest degree of intelligence at a given cost or size.
The same problem exists for x86. Is $program built for x86-64 with SSE2? AVX2? AVX512? (I chose those three because they are programmer-visible. Programmers have to use intrinsics to exploit those ISA extensions effectively.)
For RISC-V the questions to ask are similar: Is this built for RVA20? Or RVA23? (The big feature of RVA23 is the Vector extension, again something that is programmer-visible)
Embedded RISC-V programmers will have to ask a lot more questions. But for most programmers the whole fragmentation thing is simply a giant meme repeated ad nauseam.
Intel has done such a good job keeping AVX512 support away from reaching ubiquitous adoption, it's insane. There are so many useful instructions in AVX512 which are just missing from AVX/AVX2 that you can't assume exist, even on modern CPUs, because Intel can't get their shit together.
The core problem was tying instructions to bit width. But I'm actually surprised that they didn't add AVX512 support through double pumped 256-bit operations like AMD did for a while.
Capitalists don't believe in silly things like public infrastructure that might make the world better for everyone instead of making the wealthiest people wealthier. Reducing profit to save lives is socialism.
Of course there will always be sound rationale: "Don't they deserve to profit? These services cost money, so who will pay for them? And who are you to judge, for how many lives have you saved today? Aren't you just jealous?" And so on, for the love of wallowing in societal decay, happy to turn a blind eye to the suffering because common experiences feel remote when they happen to someone else, and we forget the connection to our political choices when they happen to us too.
There's no money left to spend on health because the yachts and jets can't buy themselves yet. Your life and health and wellbeing have no value if someone else can't get rich from them. So, no, no infrastructure for public good, only hundreds of millions of dollars per day to bomb Iran to distract from the Epstein files. And haven't you seen? The stock market is flying high, which is basically just as good as taking care of the poor.
If someone uses LLM to help wording their idea in a way that's easier to understand, I am fine with it as long as they don't copy-paste it verbatim into their blogpost.
I am getting a little ranty here, but the usual LLM-isms like "not this, but that" or "that's the real footnote here" simply add zero value to the text, and I can't imagine anyone who has proofread the LLM-generated text even once will opt to keep those sloppy paragraphs. If I see them, I can't help but to think that the author had oneshotted the text. And it strongly suggests to me that the author has put zero effort into their work.
reply