LLM Voice
martinfowler.com
martinfowler.com
I find the same thing with written results out of AI "research", it is a slog to read through it and most of it can be skipped even if it can sometimes be useful its an unenjoyable read.
Maybe that's the key ingredient that's missing from LLM-generated text. Even when there is actual content there, there's this lack of surprise / uncertainty in its delivery that we require to stay engaged. We may not know what will be communicated, but knowing exactly how it will be communicated just kills our innate interest.
Communication expresses an understanding. LLMs arent communicating.
LLMs generate code well enough. Code has no natural language semantics. Programming's condition of satisfaction is that the machine executes correctly. The condition of satisfaction of communication is that the communication expresses what there is to be communicated.
In the limit where an LLM is trained to repoduce an exmplar of communication, it seems as if it does so. In that limit, LLMs can reporduce anything.
If frontier labs wanted to improve "LLMs for communication" they would pay for reasoning traces to be produced which reproduced the thought process of writers/speakers thinking about their audiences. Rarely do such traces exist in prior datasets .
While I don't like LLM voice either, I'd strongly disagree with this. LLM writing can be very tight these days -- sentences do not "go on" and are not "joined irrelevantly". I'm not sure when you last used an LLM, but it must have been a very long time ago; this hasn't been the case since late 2024 for frontier models, in my experience.
Obviously there's no underlying consciousness behind it all, but the verisimilitude is good enough for stimulating conversations on the recent research literature in my field, which I appreciate. For me, at least, the communication level is absolutely fine.
(YMMV, of course, and perhaps my bar for this is lower than yours.)
They seem to be getting better at the O(paragraph) level reporting on technical updates.
They're not yet at O(pages), and certainly no where near O(book).
And the discipline gets attacked even more when all you read is LLM text. For example in code reviews there is no noticeable value in answering LLM generated review comments. I remember taking great care answering human-written reviews comments with much thought to either further build on their suggestions or to forward knowledge about the system when I had to defend a particular change.
Now, some bot will read my carefully written response and forgets about it. There is no value anymore.
No wonder that our pull request comments feel like the dead internet theory right now.
I am absolutely pro AI assisted coding. But communication between humans HAS to stay human. I am currently experiencing the opposite and it’s bad.
While I think your argument has merit, I would be careful to discount any human skill based on people being too lazy to do it.
But I don’t know where to draw the line yet. It’s difficult to decide what parts you are offloading and what parts you keep for your own brain. Because theoretically you can offload nearly everything when you get a well-defined ticket and just give it to an agent.
Maybe it’s also just me who is having a hard time staring at LLM generated texts all day and missing the past brainstorming, pair programming or “just talking about the project” sessions with colleagues and managers.
I fear that we are moving in the direction of silos again with lots of gate keeping as everyone has their own little rubber duck to talk to now.
Systems from today (if not worked on with discipline) will be no more than an agglomeration of features glued together. But as you said, guess our fight to fight will be to be that disciplined person.
This is the worst part of fully AI-generated code and automated reviews. It's so unnatural, and slightly anti-social, yet I can't figure out what the solution is when you have to ship features at the pace expected the average SWE nowadays.
And because we were talking to colleagues we wanted to make sure that we don’t waste their time. So we kept PR descriptions precise and understandable, we created small reviewable PRs and we wrote review comments that didn’t span half of your page.
Everything felt like it had value to someone.
But I agree with the current pace push it’s nearly impossible to do it any different. And honestly I don’t understand the rush. At least at our company I can’t discern any speed-up yet. Yes, more lines of code are written but features still take very long. It often just moves the work to a different stage in the process.
I am writing too much but I can’t stop thinking about all of this day and night. Sometimes even questioning whether the standards I want to live up to are maybe set too high for this day and age…
s/time/effort
I find that it often isn't just too low, but also too high at the same time. It has a lot of information that is irrelevant, burying far too little relevant information for the work needed to extract it. I wonder if this is a side effect of attention mechanism, where the AI's ability to track information in a document is far less hampered by other information, leading to it organizing information to be both dense per word but any specific concept to be sparsely spread throughout the writing. I also wonder if this is why the order feels off, because the AI doesn't need for the information to be well ordered to the same extent that a human does, so the order of storing information is often overlooked when storing it into a document (and even information that needs to be ordered might have the ordering information separate from how it is ordered in the document).
Whereas any human would say "I think your file format is missing a key/value pair for foo'ing. Let's trace through the code, and see if/where m_foo is set...."
It happens everywhere. Docstring type comments for a function endlessly just lists information about specific variables, implementation details, etc, and often never really even says what the function does. Then maybe a comment in the body might actually describe what it is for.
These days I've learned when facing these walls of text to check the last sentence first; it is often enough to provide the context for the text dump.
I have instructions to this effect in my claude.md. It doesn't help much on responses, but it give me a way to quickly say "reread claude.md and then rewrite that last prompt" and that can tend to go better.
Someone else mentioned model of mind while writing. The LLM has no model of mind; it does not consider how to present information, what you might already know or not know. We work in a huge mono-repo - it'll blithely just start listing variable names and I have no idea what file that is in, what it is referring to, how it interacts with 5 other classes it hasn't mentioned, or mentions out of order. It doesn't matter to the LLM. It can hold millions of facts in 'mind' at once; humans stall out at 5-7. If I replay saying it is leaving out info, many times it turned out it did dump all the facts out, spread across multiple prompts, often with key facts interspersed with all the vagueisms. No human mind can cope.
For foreign text, sure. But I never was never able to read New York Times articles about some invention that started with Lucy leaving her home at seven o'clock.
If you have a specific question, reading LLM text can be more interesting than anything in the world. Even Wikipedia.
I've noticed this as well. When I get a bot to generate something I am happy to read it (although I often skim read it). But god forbid I have to read someone's LLM generated report! Come to think of it, if someone marks the text as LLM generated, especially after they have plonked a hand typed summary or sentence at the top, I am quite happy to read it. Maybe my beef is being expected to read LLM text the same way I would read high effort human prose? I certainly read LLM text differently to human text - you know the first sentence is going to glaze you and the second paragraph will just repeat the first paragraph so you can skim those annoyances.
I came to think of it, LLM output is not worth to be read like human written text. It's best to just skim it and move on. The problem is when this LLM output finds its ways everywhere on the internet. Being able to skim text requires knowing the context beforehand and knowing that you're dealing with LLM generated output. If LLM output that is published was labeled as such it'd probably make us less frustrated.
These aren’t deep AI hustle bro hacks, it’s literally just editing some text. If a big part of your day is reading AI-generated words, slap something together to make it do words gooder.
But what happens when everything is persistently, subtly wrong? When we cannot smoothly map the text to a simulation? An inner-sim keeps throwing errors and mis-predicting what comes next is unnerving and draining. Like talking to a human who is not currently sane.
Sometimes I imagine LLM text like a plastic animal model put out to confuse local wildlife, where it hits just enough instinctive triggers that they try to mate with it before leaving in frustrated bafflement.
______
Half-Life 2, Dr. Kleiner: "Never fear, Gordon, she's de-beaked and completely harmless. The worst she might do is attempt to couple with your head. Fruitlessly!"
Which requires a bunch of things an LLM lacks, such as theory of mind, knowing who it is talking to, and willingness/ability to engage in short turn conversation to assess and steer the conversation according to the feedback it is getting from the person it is talking to.
Now there's an idea. LLMs may have taken the em-dash from us, but the em-en dash is unknown to Strunk & White and might be human-exclusive for some time yet.
-- - --
The Darude-Sandstorm dash
Us oldies learnt to type these with multiple dashes automatically, as unicode hadn't been invented yet.
IMO, it's still better to not let an AI critique your writing style at all, only the facts and claims you're making against its context / MCPs. Your own writing style is a reflection of your mind's thought process and, especially on forums like this one, is the only way to imbue your ideas with your personality.
Then maybe you shouldn't be making Youtube videos about it… ?
But if I'm trying to be entertained, I don't want to hear some machine generated script or even machine-checked script. It feels very impersonal, like you just want it to be correctly worded and grammatically correct, almost like you're going to be graded on it. The viewer isn't grading your grammar, they want to be entertained and see/hear you put on a performance that is your creative output.
https://news.ycombinator.com/item?id=49747070
:)
A doomer isn't skeptical; they're nearly as credulous as boosters, or maybe more so in some ways.
(But I agree with you fwiw)
Endless headlines like, "aArON SORkIn SAYs hE WiLL Never UsE ai to WRite" doesn't change that base fact. No one asked him to, other than clickbait journalists.
If you find this hard to believe, just think about relative market sizes.
Lying with statistics (well, your own guesswork not even hard figures) to say actually LLM-ese is mostly confined to TPS reports where they're not hurting anyone is in conflict with our own experiences being subjected to LLM spam/slop every day.
(Not to mention that many of the same frustrations wading through a massive volume of tediously long, low SNR LLM also apply to the workplace where the unfortunate recipient of those "TPS reports" now has to wade through all that filler to find the occasional important detail. While the sender pats themselves on the back for 10x-ing their productivity, by creating more work for their coworkers)
The fact that AI is quietly automating TPS reports in the background and slowly uplifting productivity metrics making things cheaper for us all is not immediately obvious.
And therefore…gentle reminder.
I’m happy to hear of design ideas to reduce slop on the internet that doesn’t involve a direct attack on the underlying technology.
If you built a model only based on long form journalism and The New Yorker voice, including both the template structure of the article and the grammar and syntax, the model would replicate the patterns it is built upon. You could do the same for high vocabulary rap.
When you stuff all of English or all of human thought into it with somewhat equal weight, including very technical writing the sum is a lack of style and elegance and personality.
The second problem is that writing often doesn’t happen linearly. You write bookends and fill in, edit and revise. You sort of know where you’re going and can then make decisions how to get there. Language models may benefit from being redesigned to perform passes. Select a template of some sort, fill in the mad lib, then run a detail pass, then run a persuasion / emotion pass, then run a style pass. Current model writing very much sounds like I would expect autocomplete to.
I suspect they’re leaning heavily on some kind of auto verifier, and that’s why the writing sucks, because good writing is so subjective.
And they’re probably nerfing what it can get from training as they’ve lost lawsuits for scraping copyrighted work, so they’ll purposefully avoid anything that references a specific writers style, etc.
That is, the original training on massive text corpora seems to produce an average of human writing, but that average looks pretty average. It doesn't feel like LLM slop. But the instruction tuning on specific tasks (answer this prompt, summarize this text, do this math problem), which turns text predictors into chatbots, seems to create strange style features as a side effect. And because that tuning step depends on company-specific training data (rather than just a crawl of the entire Internet), each company's LLM has a different style.
It's not that difficult to measure this. Linguists have had measures of style variation for a long time, though NLP and AI researchers seem to have forgotten. We've generated parallel human and AI corpora to measure it, and can show specifically which writing features change: https://arxiv.org/abs/2410.16107
Currently we're working on tweaking the instruction tuning process to see if we can pin down which steps and what types of tasks tend to produce the weird style features.
Though there evidently was an attempt to make GPT-5 sound more neutral, but in a literal sense: the emotional content of its text was flattened out. Suddenly it wasn't nearly as sycophantic or enthusiastic, and wrote much more dry, neutral text.
I think it's more accurate to say the text is not regressing to a mean, but to an artificially constructed style defined by the training process. And the AI companies have only partial control over that style, so weird things emerge from it.
It's not just predicting content but also the form/shape of the response, highly shaped by RLHF preferences, which seems to encourage longer form responses which need padding with filler to generate the "preferred" amount of verbiage.
It seems most of these models have been trained to interpret questions as wanting a long-form authoritative/comprehensive-sounding response (even when midwit filler is all it has to offer), rather than initiating a short-turn conversation which would better allow you to steer the conversation. Apparently Muse is much better in this regard, but so far I've resisted signing up for any Meta/FaceBook service, and not inclined to break this winning streak!
The main issues seem to be glazing that sibling mentions and also it seems to play a bit loser with facts and reality and makes up more stuff.
1: PDF warning https://www.cell.com/current-biology/pdfExtended/S0960-9822(...
I think both music and humor are both exploiting the same evolutionary essentials of human nature - to find comfort/safety in predictability, and excitement/danger in unpredictability. Laughter would seem to be a social thing - perhaps it helps develop a theory of mind in signalling that you get the joke - that the punchline is indeed unexpected to you as well ? And/or perhaps it helps draw others attention to an unexpected situation with learning potential.
In other words, it isn't just verbiage that has a "preferred" quantity, but also higher-level language features, and the appropriate contexts for those features feel like the sort of thing that will be extraordinarily difficult to replicate without a theory of mind.
edit after coming back and re-reading: the "in other words," and rephrase is because I'm trying to explain something complex enough that I'm not even sure I fully understand it, so I try typing out multiple explanations and leave the ones that make the most sense to me. This might be idiosyncratic to my writing and not a significant feature of training corpuses, but even if it wasn't it's a second example of the sort of thing I'd suspect LLMs to be bad at using appropriately.
Not to pick on GenHTTP Lambda, especially because I think it's someone's fun side project, but I was just trying to read though their docs and I was finding it super difficult. I kept getting distracted thinking "why the hell did it write it _that_ way?"
Like, the first sentence of "The control center" section [1] is the exact sort of weird negation that you're describing:
> The editor link opens a control center rather than a text box: most of the code here is written by agents, so the first thing on the screen is how your lambda is doing.
I can't help but think it generated the first round of the page with a text box to paste the code into, got told "no, I want a control center," and then this ended up in the docs.
I also think the phrasing of that sentence (and many others in those docs) is just super weird and unnatural - e.g. "so the first thing on the screen is how your lambda is doing." It also feels like it mirrors a prompt - like, "no, it should be a control center, not a text box, and the first thing the user should see is how their lambda is doing." But in docs I'd never expect it to be written "the status of your lambda" or "the health of your lambda." And what does the code being written by agents have to do with it?
A short response that's wrong is going to be punished, whereas a long answer that contains the correct information, and then a whole bunch of other stuff surrounding that info, is more likely to be rewarded. Even if you discount the reward by answer length, when there's ambiguity involved it's still going to be a better strategy on average to hedge by saying too much vs risking saying too little.
In that case, model will in fact strive to keep their writing as close to interesting writing as possible.
The "quality/goodness" (likability) of output is only selected for to extent that it also reflects RLHF preferences originating from low paid contractors who are scoring based on "looks good to me", not based on Pulitzer prize prose potential.
To explore a false hypothetical, suppose an LLM was only trained on the output of a single popular author, then you'd certainly expect it to reproduce that author's style, but it would not be "upredictably interesting" but rather predictably quirky, reproducing the same "surprises" over and over so that they no longer are surprising and therefore interesting.
To continue the hypothetical, if you trained an LLM exclusively on the output of multiple good authors, then rather than get the best of each you'd get a statistical blend (this is how training works) that emphasized what they had in common rather than the individual quirks that made them interesting.
There is a fairly common misperception that "AI" (LLMs) is always trying to do it's best, where "best" has been anthropomorphized to have the human sense rather than the LLM sense of "the most statistically accurate continuation prediction" it can make. For example, if you give an LLM the moves of a partial chess game and ask it to predict next moves for one of the players, it will NOT play to win, but will rather play to predict how the game would have continued, which means continuing at the level (ELO rating) and style of the player who had made those moves.
Of course you can ask an LLM to play to win, just as you can ask it to write in a style that is, say, "unpredictable and interesting", but it can't escape itself - you will necessarily get predictable unpredictability.
Your chess game example misses the point. If you give an LLM the moves of a partial chess game and ask it to predict the moves of the players, it will predict how the game would have continued, even if it is low level players. But if you ask it how Magnus Carlsen would win the game, it will likely come up with a sequence that can beat most human players, or possibly just decide to invoke a chess solver, which will most definitely win.
The key for your prompts and thinking loops is you need to indicate proper goals. "Write the a story about 1775 England." is different then setting a goal for it to write the a fantastic storyin the style of Dickens about England in 1775.
Regarding the ML portion; what you've written about a 'statistical blend' is a little bit coarse. That would be what you might have gotten from a 2 layer neural net or a regression in 1999; the neural networks today operating with attention or an incredibly high number of parameters are non-linear and non-interpretable, that's the entire point. This isn't a regression of simple parameters, it's all interactions of all parameters deeply netted. Model interpretability vs. Model Prediction power are generally inversely related.
The model can in fact escape itself in that it has many, many, many more parameters than a simple "predictable unpredictability" and that is exactly the point of it.
If you ask an LLM to "win like Magnus" it can't because, as above, it is a feed-forward auto-regressive predictor. Do you realize how much different this is to how something like Stockfish (that can play at, and far beyond, Magnus' level) works?
A big LLM is still an LLM - the architecture and operation doesn't change. It just has more capacity to learn, both to learn deep predictive patterns and to memorize factual knowledge.
Does it not occur to you that your beliefs of what LLMs are capable of (generating amazing prose) is totally at odds with what we see them doing? Where is all the amazing prose?
With a single, generalized prompt ("produce something unpredictable"), ChatGPT responded with:
> 00:— / ANOMALOUS RESIDUE > Vekh. > Ìssara. > Nnnn—tq—èl. > > A temporary thickness enters the meat of the signal. > > Not rain. Not a substance with permission to fall. > > UL / 9 / ferrous / ∂ > The measurements are eating each other. > > Someone has left an entire childhood in the wrong unit of pressure.
That is not predictable to me, at least. It could be boring, but I think that is mainly a product of me knowing where the text is from. LLM-generated text is inherently boring, to me, but that has nothing to do with predictability.
Nicely done.
> how awful everything is to read lately > flooded with AI generated content, all of it terrible > everything sounding identical and everything being a rambling unconcise mess.
I can see the benefits of AI and enjoy using the technology - but, like many have said, it's wreaking havoc on the "content" that's out there nowadays. Gotta take some bad with the good, I suppose.
I love computers and the internet, and the internet has ruined a lot of people and things around me.
It's possible to be interested in a new technology and enjoy its leverage while still being extremely apprehensive about how it intersects with game theory and politics. The race to the bottom of the barrel keeps coming for the jugular of every cool new thing.
You often have to make compromises somewhere in the hierarchy, especially in late-stage capitalism where even access to basic needs comes at a moral cost, and as such those compromises sometimes come with a moral duty. I am writing this via a computer built from limited natural resources, some of which have been obtained through what I see as slave labor. I don't eat meat and still have to acknowledge the harm that clearing and maintaining farmland does to wildlife, even if the result could theoretically be a net gain.
When you start viewing situations through the lens of game theory, you realise some compromises are inevitable for the rational individual. But you must acknowledge the individual is not the only concern. As such, I use AI as part of my work, but I opposed the local buildout of a large data center in my community and tried in vain to make the person responsible for the deal understand the issues it could introduce. This isn't NIMBYism, we need a halt on all buildouts until fundamental environmental and accountability concerns are addressed.
And many other aspects of my existence and quality of life, poor and underprivileged as I was growing up, have depended upon US imperialism. I have also produced a ton of waste just existing and owning pets. It's abhorrent.
I continue to ask myself how I can fulfill the moral duty which I have accrued and reduce my dependence on exploitative economic systems, but I also recognize that AI is the new social leverage and we only have so much time to use it to our advantage before it becomes institutionalized and wielded against the people. I also might only find future income stability by evolving with my industry.
Life is complicated.
I feel like in my life all I ever see is people mortgaging the future for today though. Hey how about that record storm in California?
The AI situation is different. The technology is obviously extraordinarily powerful but the benefits to humanity are still nebulous and there are many very obvious harms like the one we're discussing. So to describe yourself as "aggressively pro AI" you must believe that future, as-yet unrealized benefits will outweigh the harms. I don't think that's a crazy thing to believe but it deserves scrutiny and demands elaboration.
These challenges never go anywhere because it’s essentially a religious conviction. We’re likely in sort of uncharted territory on a macro scale and I reckon people a) either acknowledge that or refuse, and b), either trust or distrust the tech world’s stewardship of this technology. There’s a hundred bazillion data points, and I’m pretty pessimistic about the tech scene at large right now, but I think reasonable people can disagree on this without being factually or logically wrong. But true zealots in the midst of such uncertainty are driven by faith, which can certainly be deftly swayed over time, but definitely can’t be conquered in a debate.
Well, it's quite simple if you think about it. Intelligence is all about awareness of the imminence of death and the finite amount of time given to any of us. So, we humans intrinsically evolved to make the most of it. The majority of the tools we create have the ultimate purpose - to save time. Even money is nothing but a way to cut some time. It's not really about time but experiences - we want to have as many and as rich as we can get, and we innately try to cut time somewhere else, so we can use what we saved for better experiences.
Therefore, inventing cars was an easily predictable event. There's not a single hypothetical timeline or dimension where humans don't end up with cars - to transport things and to travel faster. Similarly, it was always simply inevitable - we would have invented something like LLMs, sooner or later.
Cars required significant sacrifices - they are absolutely deadly machines, they kill thousands of people every single day, kill wildlife, endager our own existence, and we're kind of okay with that. I can absolutely assure you, if someone tomorrow invents a gravitational jump travel that allows moving things between continents in minutes, even with a success rate as low as 10%, we would still be using it.
So, yes. We can't un-invent cars and AI, the only way to move forward is to adjust to this new reality and extract the best possible experiences, preferably by not robbing other people of their - i.e., enjoy life while it lasts, it still remains finite.
You lost me there. Children are pretty intelligent and have no concept of time. I don't see the relation you find obvious and the rest of your points similarly don't follow.
> Therefore, inventing cars was an easily predictable event.
This feels like a benefit of hindsight unless you can call out other obvious and imminent (but not yet existing) inventions.
I don't blame you for immediately rejecting the rest of it. But do hear me out: the point about the inevitability of invention does have real support. Historians call it multiple discovery - tons of things have been invented independently - calculus, the telephone, the jet engine, to name a few.
Neural nets, GPUs, and the scale of Internet text made something like LLMs close to inevitable. Of course, we could debate forever whether any of these on its own was ever inevitable. The point is - society tolerates any cost of a useful invention for the benefit of saving time. So it's reasonable to assume this will be the case with AI as well, which is really happening as we speak.
But, that comes with with the same kind of cultural acknowledgement, as seen with the characters in Star Trek, that such algorithmically generated material is an experimental diversion, and inherently not worth the same respect or focus as experiencing something authored with intentionality by a person.
Sure, but it didn't kill human culture. As writing didn't either. Nor will AI do it. Technology changes, human culture does the same.
It sounds silly! That's because they are completely divorced from the human experience, and they have enough power to really make us all suffer.
But that being said. I don't think that our need to push back on that particular use of AI, has anything to do with the need to push back on AI itself. Because that's not the only, or to me even the main, point of (generative) AI.
(It's just that "generating wholesale slop that people end up seeing" is the usage-profile that the AI companies can most-easily intermediate and make money on; and it's also the usage-profile that makes it most obvious that "AI is happening", and so markets AI to people not already using AI. So this is the usage profile that all the AI companies benefit from everyone focusing the conversaton around AI on!)
I think a lot of people have a failure of imagination when they think about the upside of "good" generative AI. Which to me is, specifically, the places where generative AI can be better put to use as a tool in a workflow for human creative labor to pass through, rather than being something that substitutes for said human labor.
The programming example might be the most obvious to people here: generative AI lets you write e.g. a hairy SQL statement in the form of a very long natural-language description of the query, or an example of how you'd do it in an actual imperative/functional programming language, and have that transformed into SQL grammar. So you don't have to try (and fail) to remember the arcana of SQL (e.g. how to use things like correlated subqueries, or window functions with explicit windows, or CUBE), always eventually resorting to looking it up anyway. (This is really a condemnation of SQL for having specific one-off syntax sugar for every single thing it wants to encode, rather than a uniform syntax that can fit in your head / trigger IDE auto-suggest.)
But consider Photoshop's "generative fill" tool. Yes, if you're selecting entire huge regions of the image and running that tool on them, it's going to be "doing original work" and the results will be awful. But that's not the use-case the tool was designed for. It was designed to be a better version of content-aware fill, i.e. PatchMatch. If you select a dust speck or tear on a scan of an old photo and run the generative-fill tool over it, the result will usually just be a better version of what you'd have got by running the content-aware fill tool over it. (And if it's not, you don't need to try to fight the tool; you just drop back to using the other tools in the toolbox. Not every tool is applicable in every situation, and that's okay.)
Or consider that newer versions of Melodyne use generative-AI models to better separate audio stems into their own tracks. The input here is your own manually-performed track. The output here is also your own track. Despite being a generative model, the AI here is only generating an internal byproduct that gets discarded (hypotheses for timbre- and acoustic-matched reference instrument performances) to aid the classifier and reduce stemming error/residue.
(And apparently, the compressed-sensing models used by MRI scanners are getting approximately the same benefit by approximately the same means. Less familiar with this, but it's probably the most important benefit here since improving the fidelity of MRI reconstruction means doctors can be more sure that an opacity is really "something" rather than an artifact, and so will have more confidence to diagnose real problems earlier.)
Or consider generative AI used for automatic human-language translation (where it captures much more of the speaker's original meaning and nuance than previous translation models, at the expense of giving the result "LLM voice.") As AI improves at this task, presumably there will be less "LLM voice" and the result will just sound like what the person themselves would have said if they were bilingual, because... that's the task.
In none of these cases is generative AI "producing content." In these cases, generative AI is subsuming the role not of human labor, but of tools that were already used: of "by-the-book" approaches, of heuristic algorithms, of weaker ML models relied upon by heuristic algorithms, etc. But all of these cases benefit from generative AI being improved.
And, to restate my premise: there are many tasks just like these, where generative AI currently has no place (because replacing or "enhancing" any tool in the task's workflow with gAI would currently make the results worse), where you can expect to see gAI eventually replace or "enhance" some tool in the workflow, once gAI model companies turn their attention and benchmarks to trying to optimize for performance in those sorts of tasks.
These are ultimately all software systems where updates can be pushed out with the press of a button. We just found an amazing new tool in the LLM and people are trying out all kinds of ways of using it, many of them dumb or stupid, this is discovery that has to be done, we'll keep doing it and pushing updates.
It doesn't take rose colored glasses really to infer that over time we'll cull the dumbest uses and extrapolate on the best ones. There's a huge difference between using a generic chatbot, even one driven by a SOTA model, vs targeting a specific business task with good guardrails, skills, and tool interfaces that call out to real business logic when the model isn't the right thing. The latter will emerge more and more over time.
I don't think so. AI is broad, and he's against low quality output enabled by people taking shortcuts using AI.
Using AI doesn't make your output worse, that's almost entirely down to lazyness and a lack of skill. Or people are just choosing to spend less time on the same piece of work instead of doing additional iteration and experimentation (which AI makes a lot more doable).
It makes no sense on being "aggressively pro AI" without any payment. It is like being an unpaid shill.
It's just that now, it's gone from practically fantasy to hard sci-fi.
AI has made it even worse and lots of content on the web, for both these reasons, is not worth you time (now wonder users don't read most of the text on our apps!)
Soulless Enshitificatoon Oblivion now exercises a huge weight on societies to such an extent, it's ridiculous.
In AI-generated entertainment, the names "Marcus", "Chloe", "Sterling", "Meridian", and "Apex" keep appearing. Those aren't common names, but LLMs seem to prefer them.
For me it's mostly the latter, which is a natural consequence of LLM writing. When the (time) cost-per-word goes down, there is less incentive to keep signal-to-noise ratio high, so it ends up verbose and bloated. I also believe that this is (at least partially) upstream from everything sounding identical: take two articles on two different subjects - when SNR is high, there isn't really much room for them to become similar (The ingredient list for a risotto will never read the same as the ingredient list for pizza), but when 80% of them is fluff, they can blend together more easily (To take the recipe comparison, this would be the background/story part above the recipe), especially when written by the same LLM
I've noticed this a lot in conservative spaces especially, when they say "I love Trump, but..." -- why the need to say how much you love AI before criticizing it? Who are you trying to convince?
I use LLMs constantly for work and they have genuinely improved my life. But I absolutely despise generated "content". Text, audio, video, images -- for the reasons you describe.
It's ok to have a nuanced and multi-faceted opinion and approach to this and not have to call yourself a raving fanboy.
My parents live on a rural property, and I'll help them do research on things like livestock practices or occasionally "prepper" related things to give them a sanity check before they spend money. I have to weed through a dozen or more obvious ChatGPT-created pages that are the first page or two of results on every search engine before I get to anything a human has touched. The vast majority of the LLM-created pages are either massively verbose and vague, or often wrong. But they're getting clicks. Very often, when asking an LLM the same question and to give citations, I'll get a combination of useless, irrelevant sites with no connection to the material, maybe one of these LLM-generated sites, and maybe one online forum.
I'm not aggressively pro-AI, I'm in the middle, but stuff like this has me worried that we're looking at the equivalent of the Dark Ages all over again. But instead of simply no one bothering to write anything down as was the case in the early Middle Ages, it's going to be the task of searching for real human work and thought in a sea of aggressively spammed LLM-generated slop pages.
"You can be pro AI productivity but anti-AI creativity. Automating creativity is the death of art."
Meaning you reject the creative value of AI but values it's ability to drive a car safely, or write code for your job, or turn off the lights when you leave the room. objective goal oriented task completion great.
But AI draws your a picture? What's the point. AI writes you a song? No feeling. AI creativity isn't interesting.
There isn’t any doubt that AI can be a useful tool. So what you’re essentially saying is that we should avoid using a useful tool because many people, without any relevant training, use it badly.
By that standard, you should not be pro-car, pro-airplane, pro-boat, etc. etc.
What you're doing is engaging in major denial in response to a change that, for whatever reason, you're unable to handle.
Max Planck's famous quote is relevant here: "A new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it." Much the same often goes for new technologies.
Of course it's not purely an age issue; plenty of older people are embracing AI. But the kind of people who are completely unwilling to entertain the idea that it could be a useful tool fit Planck's quote perfectly.
This reminds me of the people who think they're having economic trouble because immigrants are taking their jobs.
You'd have to tell me what you see when you "look around", because it's clearly not what I see. All the problems I see come from people doing things that they shouldn't be allowed to do, that should be taxed, regulated, etc. I see monopolies that should be broken up, money being used to drown out opposition with the backing of law, financial scams of all kinds propping up the supposedly most valuable companies in the world. The idea that "AI" is the bogeyman that's somehow causing any of this is drivel.
But you write one-sentence comments that don't actually describe your position, so it's impossible to know what you're actually trying to say. Perhaps you should ask an LLM to help articulate your ideas, if you want people to take you seriously. And then you'd find that AI can actually be a force for good.
I'm not using an LLM to write comments, because I'm capable of expressing my thoughts in writing on my own. Which goes back to what I wrote earlier - the issue is how people use AI. AIs aren't forcing people to abdicate their intellectual abilities. And in a case like yours, where expressing yourself is not your strong suit, they can help.
> And in a case like yours, where expressing yourself is not your strong suit, they can help.
Do you even read the shit you type? Or copy and paste because you’re a sad little ai driver? Of course I’m not interacting with you in this conversation. You’re a fucking pompous retard.
OP’s linked article mentions the importance for them of reading their writing aloud. Humans construct sentences with a (mostly unconscious) sonic influence. How things sound really matters.
My theory is that LLM prose sounds slop-like because the base unit for an LLM is not a single sound.
Thus, perhaps, you get the uncanny, meandering, hollow-feeling outputs, which feel vapid to human readers, without the reader really knowing quite why.
I think this is also part of the reason why LLMs are rewarded for and thus more strongly exhibit their faux-punchy constructions, such as the infamous “it’s not [X], it’s [Y].” The transformer model can, with current tokenization, articulate some features of “good writing”, but not quite all.
I think there are a number of fairly simple SOLNs to this (or ways to test this hypothesis) without making tokenizers purely syllabic across the board, which would probably be a mistake.
Whether prose output quality even matters for models which are not explicitly trained for creative writing is highly debatable, though.
I also try to keep the llm-enhanced (or even fully llm-generated) content separate from my human writing - because it is true that the failure modes are different. But think LLM writing might be useful even if it is not consumed directly - it can enhance LLM answering your questions.
(this isn't a new observation / thing, see https://randsinrepose.com/archives/nadd/ originally from somewhere in the 2000s)
But now with AI it's easy to generate a lot of prose. I'm thinking for example of the merge request descriptions my colleague submits. And here's the thing, it's not even bad or incorrect - I read it thinking, if I were to really sit down for it, get into The Zone etc, I could crank this out myself.
But I don't because honestly the MR is relatively trivial and can be described in just a sentence or two. Style and accuracy aside, my current take is that LLMs write things that humans wouldn't bother with, because it's too much effort and there's not enough value.
I mean it's an MR for an enterprise library project, the chances that anyone will ever read that MR description again once it's merged is close to zero.
That said: LLMs themselves don't get that, they will read everything you tell them to with the same amount of attention. Granted, their context windows are limited (and performance declines the more you put in it), so they too will use some tools to skim and pick out the important parts, but I think that's a temporary thing. The huge amount of LLM prose will end up being consumed by other LLMs. I don't know enough about the subject whether that's a good or a bad thing though, right now I think it just adds to the noise until it becomes noise in, noise out.
But then
> I've also noticed how awful everything is to read lately
> A site I read stories on has been flooded with AI generated content, all of it terrible
I'm not taking either side, but I'm curious about how you reconcile being aggressively pro something that, in your own estimation, is making everything terrible.
Reading some AI text feels like looking at a Michael Murphy sculpture - makes perfect sense, until you start to look closely at the text, and gaps appear - and you start to realize there's no there, there.
I know this because the topic is usually an obscure issue, and I have already gone back-and-forth with multiple LLMs to see where their boundary of knowledge is on it, and it will often include hallucinations along the way (e.g. pointing to old/nonexistent settings pathways)
So when the first reply from the official 'support' channel contains the exact same nonexistent settings pathways, it is easy to conclude they are sending me the LLM output without doing the verification steps that I have already done on my end. When I point out what doesn't exist, the reply is always a similar flavor of the usual LLM "you're absolutely right" response after being corrected.
I'm all for having these bots process inquiries if it means a speedier resolution, but, please, anyone out there using these things, turn off the simulated sympathy. It's infuriating how condescending it comes across.
(yes I understand this is likely not legal or practicable)
The bots are specifically trained to do this because lots of people are dumb enough to prefer it that way.
It's something ELIZA could do in the 60s, and that worked unsettlingly well for an unsettling fraction of people who tried it.
This bubble is bonkers, if our Jetbrains contact did the same we would get very mad and yet we spend significantly less with them AND we don't send them all our data.
Even if you don't think it's a problem, even if you believe that it's just different than human writing, it seems the majority of "regular" people can tell when something is LLM-written and find it distasteful or unpleasant to read, at best. That alone should be more than enough incentive for frontier labs to train this out of their state-of-the-art models, if only they could. The fact that they haven't suggests that they cannot, despite all the claimed advances in model analysis and "explainability."
So diverting significant resources is just going to harm the main goal and kind of suck at the side goal too.
4.5 is a great example of this: it wasn't perfect but it was better than every 5 family model at things like tone and writing, and despite being priced extremely ruthlessly and already finished training, they couldn't afford to tie up the resources to host it.
I always had a high faculty for language, and when I was young, even very young, I was able to speak in full sentences much like I'm doing now. What I didn't have was context. I didn't know how to relate to people socially, and then later on, I wasn't as experienced an engineer as my peers, but I was trying to act adult and mature in both ways.
So I tried to write in ways that were catchy and unnecessarily complex, because I thought that conveyed sophistication and intelligence, and that I knew what I was on about more than I did. It's the kind of writing that gets you through a 30-page final paper for a class you don't really like, or a research project you didn't finish. The term I often use when complaining about my colleagues' LLM output is "flashy but meaningless."
As my confidence grew, I started writing in simpler (though still not the most concise, that's just not who I am!) ways, because it became easier to convey ideas simply in a way that made it obvious I understood them.
When you think about what LLMs are - word machines without (or with very limited) context - you can see why they write the way they do. They are always acting; trying to replicate a human contextual experience with only weights and tokens to carry them through it. That fact is part of why I am skeptical of their ability to suprass humans in reasoning based on their own output.
Maybe it's like when you can just tell a picture is AI. You recognize the words you are reading, but to your brain, the way the words are put together feel wrong.
Incidentally, I was reading Harry Potter to my kids last night. While waiting for everyone to arrive, I started reading the previous chapter to those who were present. But I read the sentences backwards while using a vocal inflection of regular narration. My 6-year-old recognized that something was wrong and asked me to "stop reading funny". My 11-year-old recognized that I was reading backwards. The same 6-year-old thinks every silly photo that I show him is AI, while my older children can actually spot the AI.
A very important point, which was also made more forcefully in "Human Routers of Machine Words" by Fernando Borretti, shared on this site previously. It's good that Fowler brings up a very legitimate use case in the next paragraph, though.
Surely someone with a neuro/psych background has picked this up and developed an evidence-backed explanation.
AI does the same with responses where the cadence is some type of formula that becomes identifiable.
Each of the big models has their own but an example of the cadence is
A) Agreement or Acknowledgement with the User
B) Specific detail gets a paragraph of detail related to the previous response
C) Overview of the conversation's current state
D) Prompting or Direction for this conversation to go
E) Landing on a beat. Wrap up the vibe. Recap with a zinger.
I cannot unsee it and a few of the models are annoying. I have attempted to build out the Agents and context to avoid the 'last beat one liner' without success because after 2-3 more turns, I get "That challenge was accepted and you stepped up. You're the winner this round".
Compared with human writing, all nine frontier and near-frontier models use more inanimate subjects with animate verbs and more low-frequency n-grams, and most also use longer dependencies, more stacked nouns, and more reduced object relatives in conversation. [...] We propose and test a hypothesis as to why: while human language is optimized for successful communication, AI-generated text is increasingly optimized for other goals like math, code, and reasoning, in which there is a pressure for information density rather than successful communication.
[1] "[Robert Caro is] a guy who genuinely rarely gives interviews" https://manifold.markets/DylanBowman/who-is-dwarkeshs-upcomi...
[2] Scott Aaronson uses it four times in a short post: https://scottaaronson.blog/?p=10222
>Me, I voted straight-ticket Democrat today, as I’ve done in every election since I became eligible over a quarter-century ago (except for when I vote-swapped with someone from another state).
Or think about reading something that might be stuff you could say in your own words but not actually in your own voice. Or maybe the opposite is a better example for you? That's what some poetry is about. The gap between voice and word.
Stylometry is the study and measurement of text, so that can go some distance to quantifying the difference between voices and can help to programmatically identify a human voice but I feel that because a voice comes from a person, the recognition is also mainly a human response.
At another level its about communication, person to person with the intention and desire to communicate something. LLM text output may seem uncanny as it's hard to emulate the intention and desire and empathy present in genuine human communication.
I would be interested in reading what poets say about LLM voices. I wonder if many LLM labs employ them?
I also find LLM-voices extremely jarring, boring, and dense. And very, umm, mechanical, if you will.
But I want to make an observation.
Folks are not-Ok reading the LLM’s output for prose.
But folks are completely Ok, even downright blind, to accepting LLM-output of code.
I see that dichotomy everywhere.
LLM-driven code: great, valuable, all-ok LLM-driven prose: not ok
Understood completely that the audience for consumption is two completely different entities (machines vs humans) but something about this doesn’t sit right with me.
More particularly: if we don’t prefer LLM-prose, why should we prefer LLM-code ??
Anybody else feels that ??
There are no objective criteria that can say "this text is good" or "this text is bad". Sure, spelling and grammar errors... but we're talking about writing style, "voice", and whether the text is easy for humans to read.
Now, there are plenty of problems with LLM-generated code that can't be solved by having good tests. Is it maintainable? Is it written in good coding style? Those are harder to quantify. But since LLMs are such good mimics, if you tell an LLM to add code to your existing project, then it's usually going to do a decent job of matching your coding style. The question of maintainability still rears its head; my personal experience is that the smaller the chunk of code the LLM generated for me, the more likely it is I can understand it fully and therefore maintain it in the future. An entire vibe-coded project? Good luck maintaining that. A 100-line class with five 20-line methods with names that make sense? Yeah, I can take that and maintain it in the future, even if the LLM tool goes away.
But mostly, the difference between the acceptance of LLM text or of LLM code, IMHO, boils down to "are there objective measures that I can use to accept this"? Code has to work, which is a much higher bar than text (which just has to have correct spelling and grammar), and therefore it's easier to get the LLM to spit out decent-ish code than decent-ish text.
I had Claude build me a map tool to calculate travel times from one or two points to every street in the city; not only did it do that, it added various scenarios, speed limits, usage of bus lane toggles, different visualizations based on thresholds, and a tool to scan for optimal starting points based on a configurable grid size. etc.
This overcompleteness is a mixed blessing though, in this case it was great and convenient, but if it was e.g. documentation intended for human consumption then it wouldn't be so great, and if it was a letter written to the county that they should consider another location for a fire station it'd actually be pretty ineffective.
That being said. Yes, I hate some LLM Voice moments too and still sometimes judge people for copy and pasting, expecting others to do the comprehension when it seems they haven't done it themselves.
Duality of man.
Look at programming. What's happening there is we are having agents produce 10-100x the code to where people give up on reading it and just hit OK. Look at math. Plenty of people there want to learn from the LLMs and all these crazy new math proofs. But those are the people who will probably be out of a job soon because the money in math is in producing proofs, and that's now an LLM-exclusive job.
I think LLMs are weird in that the first order effects are those of an incredible and amazing tool that anyone can learn things from, but the second order effects are going to be the exact opposite - people giving up on bothering with ideas. Even at my job, which is one of the few that has not gone crazy with AI, anytime I ask anyone a question, all I hear is "have you asked Claude?". There's no communication anymore, no pushback, no struggle. I worry that one day we'll look around and realize we've built an entirely separate second civilization of machines and robots that acts more like a neighboring country than a tool we can use to build humanity up.
-Just my rambling thoughts, not refined by an AI.
I worry that--while they can be a strong fuzzy search tool-- their teaching ability falters because teaching involves possessing a theory-of-mind for the student and trying to figure out what will change how they think. A good teacher doesn't just present you with relevant facts, they try to figure out how you're reaching conclusions.
Note that I'm excluding propoganda-style "teaching" here, where a model just insists you're wrong and that all the cool kids are doing it etc.
(I have to admit, I truly do like some of the music, so feel free to scorn me. If you want to be infested with a vicious ear-worm check this out: https://www.youtube.com/watch?v=zONGVaONKe8&list=RDzONGVaONK...).
A writer who can't recognize LLM voice isn't a good writer, just like a writer who doesn't do research or fact checking isn't good. It has nothing to do with method, its about skill and work effort.
And of course I’ve dropped my use of the em-dash after being accused of being AI myself.
But of course: manufactuting itself creates " productive mechanism whose parts are human beings, as Adam Ferguson describes the industrial workshop: "an engine, the parts of which are men".
We should actually restructure our organizations/society in such a way that this stops being necessary. If no one reads a report, it was almost certainly a waste of effort. The only possible benefit was for the report-writer (which of course is also flushed if the "writer" uses an LLM). These things were already a plague before LLMs.
Boring reports are also important for historians, although granted, not everything is actually important in the long run.
If the solution is changing the format of the reports so they're less heinous to write and read, I'm fine with that. Let's just make it actual communication, not letters to the void.
Automating creativity is the death of art.
The typewriter was the beginning of the end of the art of handwriting.
Factories was the beginning of the end of hand-created goods.
AI has killed the art of handwritten code (which itself automated away other forms of art).
The life of a programmer these days feels very weird. "What am I good at?" and leads to even more impostor syndrome... "I seem to still be 'good' at prompting my LLMs and managing workflow design." but it's definitely a different skill/feeling than when you cranked out some awesome piece of code that did something very neat.
I felt this way a bit when I was tech middle management; should I feel like I'm getting credit for my team's work, when I had very little input aside from helping with the "unblocking" when they ran into a tough problem.
Ah good ol' serendipity. I strongly believe this too, and I think that's why older (less capable) models fared better: they hadn't fallen into the valley yet. Maybe it's not a valley at all, and is just a cliff after which there is uncanny.
When the earliest models launched I was super-excited to no longer have to write pull request descriptions and commit messages. I have done a sharp 180 on that and I'm no longer a meat proxy[1]: instead I point out any redirects and improvements I made the agent do before opening the PR (which there are still _many_ in even the most trivial tasks).
Given the amount of angst being expressed in this thread about particular LLM-isms, I wonder why people don't simply prompt against the traits they hate. Yes, it will reduce output quality very slightly in some circumstances, but if it really gets your goat like this, surely it's worth it.
I enjoy cooking from scratch, and I really don't like frozen foods, microwave/TV dinners, or even DoorDashing
but we're past complaining/arguing/shaming folks who don't cook every meal from scratch
you know what, I don't prefer eating it, but if my next door neighbor loves microwaving frozen $3 burritos, then go ahead
to some people, food is art, creativity, life to others, food is fuel
also applies to writing and thinking
You can love to cook and enjoy every second of it, but sometimes you just need fuel and that $3 microwave burrito is sufficient. Just the same, 99% of your means could be frozen TV dinners, but that doesn't mean cooking a romantic meal for Valentine's Day is off the table.
There's a weird level of fear mongering occurring around AI that it will make you incapable of independent thought if you use it. People will throw around very light research that kind of points in that direction, but they'll never highlight any of the nuances in such an argument. You can use AI and still "think" if you choose to, just like how you can avoid AI and still opt out of "thinking" if you choose to. It has less to do with AI and more to do with the context of the individual using AI.
When I come across their slop I need to remind myself to let it go, else I’ll miss out enjoying the food from that Abuela who’s been cooking longer than I’ve been alive
It’s not ok to write stories with AI and post them for others to read claiming “I did that”
Get the difference?
Some of them purchase ingredients from those companies, and then make the food using those ingredients.
A great deal of restaurants, though, simply purchase pre-made frozen foods from those distributors, re-heat, and sell it to you at a premium without disclosing while you think it was freshly made. Mac & cheese, stuffed peppers, lasagna, burritos, most 'bar snacks' you can think of, etc.
basically any food establishment that has "supply chain operation" is slop to me; I guess it's similar to my distaste in reading AI slop too
I don't fault ppl for going (and enjoying) those too- Chili's can be enjoyable after a few beers
I single out articles because I know many people consume AI video slop when they are tired and just want to disengage, like eating junk food.
happened to me once lol, will never forget that place ever again (blech)
right. he is talking about the first group. where did you get the 'shame' part?
I like your analogy. Microwaved empty calories perfectly describes llm output. You are going to pay the price sooner or later.
That is not to say that I don't think there is a place for LLM assisted writing. Writing is an exercise for the author and translates to oral argumentation so I don't think that fades away. But LLMs could do some "audience normalization". What if I could make a blogpost that is dynamic based on experience of the reader? There are lots of opportunities, but versus other fields where optimizing for laziness can be beneficial I don't think lazy writing currently has much value add.
It's your turn to host supper club this week and you serve frozen diners on your nice tableware.
Same thing with LLM-assisted writing. It’s one thing to do it for yourself, quite another to share it with other folks without warning.
Most food is literally just fuel. I think that is fine. And at some level, eating 3$ burritos for a quick lunch on the daily really makes you appreciate the depths of a home made risotto even more. :]
Add to that the fact that AI is significantly worse for you than TV dinners, and has an outsized harm to humanity, and the conclusion is plain:
AI should not be imposed on others.
Or maybe the humans will do the opposite and shift to Sardukar-style communication instead? LLMs can easily match that as well.
That style I would very much prefer to all the fluff people write today and were writing even before AI.
* Every user run becomes a memo
* Days summarize, rolled to weeks, rolled to months - all written once
* Today's runs are shortened and placed into running context past the cache break
Then while reading it before publishing it, I thought it sounded super LLM-y. I didn't rewrite it because I jammed that stuff together before hitting publish because I was in a hurry but now that I go re-read it, it's somehow worse. Dear god. It's happening.
The classic thing is this business with inanimate objects verbing metaphors or becoming other objects. It's a strangely uncanny valley, but I suppose I read so much of this with Claude that it gets stuck in my head!
It's a lot like talking to someone who doesn't know what they are talking about but are good at faking it for a while. The game works by leaning on pat phrases, circular reasoning, and hoping nobody is paying too close of attention. I think this is why it's so popular with the corporate executive class, as that has been their mode of communicating their entire careers. To them, they don't see anything wrong.
"Shitty writing existed before AI" is unfortunately an all too common rejoinder to this issue. That's not an excuse. That doesn't make it better. Just because the pointy haired boss talks that way doesn't mean we should be gleefully pouring gasoline on the fire.
I haven't really done a systematic analysis to compare the quality of fully human-written content and content generated via this kind of deep human-AI collaboration, but my intuition is that the latter is vastly superior. Perhaps AI at its best is a research tool, but a genuinely intelligent one that we need to engage in conversation with in order to use to its full potential.
The sense that I get is that as a conversation gets more in depth, the LLMs start behaving more like reasoning engines and less like statistical averaging machines. The former is vastly more useful to a person who wants to develop an idea and articulate it well.
To me, for most business and academic content, the ultimate goal is to convey information. If I can edit AI copy to legibly convey that information clearly, that's enough and I can move on.
The saying, "Don't allow perfection to be the enemy of the good," comes to mind.
And I'm not convinced an individual can accurately edit without learning just as much as if they wrote the content. I guess the question is: how hard does the person work at editing. And does the individual challenge themselves to not put out content they themselves don't understand.
We're in a transition period. And as in all transition periods, many recoil at what is new. LLM writing still feels strange. And it's understandable.
But, for better or worse, in a business and academic setting, as the years go by, most of us will shrug and move on.
As with any topic, there will always be a minority of fanatics who cling to the rejection of AI writing in all its forms. Just as some prefer Vegan leather or tabs over spaces.
But among those who get things done, most (though not all) will accept that it's not worth it to exclude AI from writing. There simply isn't sufficient ROI.
It's like trying to bring back low-end highly human employed manufacturing to wealthy industrialized nations.
That ship has sailed.
I do agree, however, that in cases where an individual is reading for pleasure -- e.g. not in a business or academic context, authors will be forced to pick their audience: those who enjoy writing, even if it smells of AI, and those who don't.
I guess to summarize, my feeling is that using LLMs to write is perfectly acceptable, and used well you'll likely learn just as much. But you must be willing to lose some of your audience. And, over time you'll lose less and less of that audience.
I turned on ChatGPT on Voice mode with no context whatsoever and just put it there in speaker in the room.
It started giggling and say "yeah" "right" and some very weird stuff. Then one of our friends reacted to that and started calling it out, then it started changing.
I'll keep it short, but basically the thing was super hilarious we had a lot of fun and at the end it triggered it safety protocols and its voice completely changed and it was acting super weird.
I know this post isn't about that, but I feel like soon we will have true AI companions in the home and around people.
It's human to do that because it's too dry and unwelcoming to talk in precise concepts disregarding any cognitive friendliness, not to mention do that all the time.
A computer programming in any high level language is concept talking all the time and only us nerds like it :D
Not even that as we form tribes with preferences to one or other computer lang.
Now we're talking less to each other and AI is going to influence us to invent a trillion things that will make understanding diverge and start to become scarce.
I have to believe that we are within a couple of years if not months of this. This problem is too well-known to not be a major focus of the frontier labs' efforts.
Not least because it’s getting worse rather than better. But also because as AI gets better at writing like humans, humans will change how they communicate to avoid being confused with it.
Which will get a lot easier once human writing is nearly extinct. Now that could happen in two years.
For example, I don't dislike the way Gemma 4 writes so much.
But I am British and Deepmind's output somehow seems more compatible with the way British people communicate (it's a very multinational team, I suspect that is an influence).
Other people pick up on Gemma's AI writing tells a little more readily than I do. I have a greater tolerance for Gemma (and Muse Glimmer) than ChatGPT or Claude (or Qwen).
The reason people notice "AI voice" is not that the AI voice is itself remarkable. It's that it appears so frequently. Every literate person spent nearly their entire life reading text written by a variety of different individuals. Dozens of different voices in your morning paper. Hundreds of different voices on every web forum. You get the idea.
Now, imagine that all of those voices were replaced by one guy (or two guys, or three guys if you're a Gemini believer). Everything in the newspaper is written by one guy. Every comment on reddit, and HN, and the NYTimes comment section, all written by the same guy. Other people tell him what they're trying to say, but he's writing all of it. Some people might even ask the guy to write in a different voice, but the voice is just that guy doing an impression of someone else, it's not actually someone else.
That guy's voice would probably start to get pretty grating. You'd start to notice every tell, every frequent word choice, every formatting quirk. You'd start to wonder if that guy was actually a good writer, or if having him write everything was just more convenient for everyone. You might even start writing like him yourself. This would be true even if that guy was an absolute virtuoso of the language and his writing was unassailable in a vacuum.
If everyone is using the same (small set of) model(s), then it doesn't matter if the model is GPT-3 or Claude Legendarium 9. Unless the operator - which in theory doesn't need to be a human - is doing something to move it off of its default voice, the outputs are going to sound the same, and you're eventually going to figure out, and get annoyed by, its tells.
Yeah. I want to be able to touch a human being (or at least have one touch me) through the words. LLMs poison that, even if a human prompted the LLM.
Translation... I think I object to translation less than Martin does. There's a human back there, and they're writing all the words, and then an LLM is translating them. The LLM may not translate them perfectly, but they aren't going to introduce "load-bearing" when it wasn't there.
The bit about using LLM when you're writing for a bureaucratic process rather than a human was interesting, at least to me. I hadn't thought of it that way before, but I think it makes sense.
Those all sound reasonable to me. What he staunchly ignores is that coding is essentially writing, but that ship has already sailed!
1. If you use the first person pronoun "I", you better make damn sure you wrote those words.
2. It's fine otherwise to use LLM words, but I beg you - DON'T SHIP YOUR FIRST DRAFT. Read it, reread it; delete words in it; replace three words with one; replace dumb or overused turns of phrase with fun metaphors; generally give it a sprinkling of joy; break up punchy sentences by making them run-on sentences to trigger your readers' OCD... you get it.
Did I mention not shipping your first draft?
I wrote down a few of my thoughts a few weeks ago at https://notes.pault.ag/why-write/ -- I find myself scanning text to see if it's LLM before I actually engage with the content.
It is getting harder to do by the day, and has slowly moved from a mental exercise (look for specific markers) into a "gut feel" based on how the pieces fit together.
Statistically even or repetitive over a chapter, chapter after chapter. So make an exaggerated point prove it come up with a conclusion and then have an exception.
Repeater 4 to 6 times a chapter in a book of 45 chapters.
So I wrote something to remove it and yes it is better but still not very good.
I created a small tool to make this easier: https://www.writelucid.cc/. It's free (bring your own API key), feedback would be appreciated.
I assume that only applies to natural language, not computer languages?
I think the key difference is that code is a tool used for a given purpose. It can be slop, or structured better or worse, but at the end of the day it exists to be executed, and its execution is either correct or not. Additionally, good code that does a given thing can only be expressed so many reasonable ways, probably. So LLM-generated code can be perfectly fine.
By contrast, human writing is about exchanging thoughts. Often those thoughts are about infinitely complicated and poorly understood systems, like other people, or emotions, or politics, so it's not a matter of exactly encoding something but attempting to describe it imperfectly. All the tiny little details encoded in how someone writes something can work to this end and are also a product of the writer's thought patterns and overall personality; this all conveys useful information to the reader. In other words, there is a much greater degree of fine detail that affects the quality of writing than of code, and LLMs suck at those fine details for most subjects that are interesting. On top of that, they don't have original thoughts.
>Like most writers, I find writing an almost-essential tool for thinking. Often it's only by writing something down, that I can really get my thoughts on a topic into coherence. ... . As more people delegate writing to LLMs they will also delegate thinking and comprehension to LLMs.
There's an analogy with LLM coding that is so screamingly obvious I'm embarrassed to point it out (changing "writer" for "programmer").
I'm curious to know if the author, given who he is, thought of it (he had to, its just too obvious) and what he thinks of it.
It is impossible to reconstruct the intent/spec from the code. More often than not the intent is not clear in the code, and one has to track down the original author to check their context and decision at the time.
As the saying goes, naming things well is a hard problem. It's kinda mystical in a way, where only by knowing an entity's true name will one understand it.
However if it's communicated between agents one may care even less.
To this point, however, writing informative articles about regulation or legal topics goes great! I don't want there to be ambiguity.
One tactic I use is to point out specific phrases as awkward, and create a tone guide both for the series and the specific article. They can mostly understand when you show them, it makes the process easier!
Here, this reader wondered whether "to" was omitted intentionally to emphasize lack of LLM involvement. Then, on a second read through to find the sentence again, I noticed the first paragraph is also missing a period on its final sentence. Which just goes to show: we can all use editors and proof readers. (According to the acknowledgements section, nine humans read drafts.) Now, I don't want to intentionally leave typos in my writing, but for me it's preferable to LLM-voice no doubt.
> But I also would prefer rather a genuine human's clunky prose to the smooth but soulless LLM.
Amen. This reminds me of a bit from Kitchen Confidential about how Anthony Bourdain would prefer a simple and plain home cooked meal over an ornate restaurant meal any day of the week. I read because I want to hear what a human has to say. It's the errors and malapropisms and style that give a unique and human voice. Doesn't this contradict my earlier statement about not leaving errors in my own writing? It turns out I'm perfectly tolerant of errors in anyone's writing but my own. (I should probably bring that up with a therapist.)
It's something that is of course fixable. Either with a lot of prompting and direction. Or by training on existing text that you wrote. To some extent some tools already do this. But of course if you only ever let AI generate what you say, how would you even develop your own voice?
On a meta point, I did write all of this myself.
If you want to personify them: it’s a million clones of the one person, who is a tolerable actor, able to give impressions of others on request, but still ultimately one person.
As far as different models are concerned, they’re disturbingly close to being the same person, but perhaps we can call them siblings, raised in the same home.
HN uses automatic classifiers. They seem pretty effective. At the same time, I think it's a loosing battle. I'd rather HN either flag for others to ignore/filter/review if they want to, and/or allow people to self-report (ala TikTok)
Eg. I wouldn't mind if a comment on HN was like:
Hey everyone, I coded this little thing:
<ai-generated> some code </ai-generated>
Then users can choose if they want to see posts with AI-generated stuff or not, or something else...
Of course there are always going to be some people trying to game the system. But what are we optimizing for? (a lot of times the 95% unnecessarily suffers because of the mitigation efforts to prevent the other 5% from abuse)
The “Simplified Technical English” fad was pretty effective, but now I just specify:
- Use short sentences and active voice.
- NEVER use the phrases "load-bearing", "seam", "unlock", "real", or “learnings”.It's infuriating to read, but even more so, it feels disrespectful. It's like they're implicitly saying, "I can't be bothered to respond to you, so just parse this AI slop".
I don't really know how to make a case against it since leadership is very pro-AI usage. Has anyone dealt with something similar and have suggestions?
1. The writing is terrible to read and wastes time.
2. The act of writing ensures understanding. If no understanding is needed, then this doesn’t need to be in Slack, Claude could be just doing it.
I really don't like this timeline ... sigh
I wonder if you took the pure speech to speech models and STT'd the output if it would remove the LLM voice? After all, can you even imagine hearing a human utter aloud some of the nonsense you see spewed by claude? Or give a longform presentation or speech in the form of the essays that are 100% LLM-written?
Maybe there's something in our human brains connecting oration and writing that influences the style of writing that pure-text engines just lack?
With much sympathy for the headline this is the central problem.
It's not easy to say what Uncanny Valley looks like, and so we're left with suspicion not certainty.
BTW, I am now 4-5 out of hundreds of my comments since AI detection became a thing, called out for having "AI" tone. I fucking hate being objectified like this. It's not true, although I understand that assertion is hard to refute or support. All I can say is, some people write like some people believe an AI writes.
And I wouldn't.
> (it's not the em-dashes, it's the load-bearing tapestries)
I suddenly feel an urge to turn this into an image prompt.
Because many good ideas are not immediately obvious good ideas, in an environment at risk of information overload, we need some way to quickly vet ideas to see whether they are worth investing the time to consider them. (It is often the case that the best ideas require considerable investment of time and effort to be judged as good, and, it's also often the case that this determination can be made only at the very end of such a process.)
“Don't judge a book by its cover,” sure, but, historically, there have been superficial markers of care and attention that we've used to these ends. These are obviously very flawed signals, but, you know, maybe if you took a lot of time to craft a message, I don't feel that bad if I waste a bunch of time trying to decode it. These markers—attention to detail (e.g., spell and grammar checking,) clear-structure (e.g., “bottom line up front,”) and typographical flourishes (e.g., smart quotes and em-dashes and extended unicode like diæreses…)—were an indication of some modicum of quality control and care. I, myself, used to brag about or show-off having set up .Xcompose keybindings (https://wiki.archlinux.org/title/Xorg/Keyboard_configuration...) which was just rare enough across the general population to be noteworthy.
All of this that was once somewhat expensive is now cheap or basically free.
Even if I make a point of going back to a pre-UTF-8 Usenet style (or even intentionally insert run-on sentences or misspellings or grammatical errors,) the cost of an LLM to style-transfer its output to match this is barely much of anything. All it means is I wrote one prompt for the first draft, and then one more prompt for the refinement.
Additionally, some LLM habits are actually good habits. Putting the bottom-line up front (in bold, even!) is a great habit in a business e-mail! Decision-makers genuinely struggle with reading through long paragraphs of detail!
Unfortunately, the problem with the LLM output is that it takes good, valuable, superficial or structural habits that it can cheaply apply, and it drapes those around inane, banal, cloying, muddy, meandering meaninglessness.
Today, we can maybe still judge a book by its cover, because even one more prompt worth of refinement is still too much effort, but I don't think that's amounts to much, since the frequency and loudness of complaints of LLM writing means that model trainers are going to train their models to adapt to whatever signals we use for quick vetting. I am skeptical whether this training will concurrently improve the novelty, innovativeness, or other intangible qualities of the actual message. (With the models, as they exist today, and as we use them today, at least in the field of software development, I see extremely poor capability at generating innovation… Personally, I don't even like the quality of the Python code they spit out, but they sure do type real fast, and, like, pejoratives aside, unexciting, mediocre, corporate-y code does have a real place in the world.)
So what happens when the economics of production in the medium significantly outpace the capabilities of production in the message? The LLM can write a comment nearly indistinguishable from any given human… except when you actually take the time to read through it, it's just inane drivel. Thoughtless and a waste-of-time. And there's basically no way to manage your investment of time and effort to tell one way or the other.
Because, for sure, some of the real bangers of comments I've posted on HN and similar forums—all my XCompose-configuration aside—have gotten relatively few upvotes!
Same thing with conference talks. The ones with half a million or a million views on YouTube… the real magical ones, I think, are the ones no one seemed to pay any attention to… e.g., https://www.youtube.com/@PyDataTV/search?query=james%20powel...
If you care about your voice, don't let LLMs write your words. But that doesn't mean you can't use AI to think, critique and draft lots of words for you. It depends on what purpose you're writing it for. If you're writing an impersonal document, like a design document, briefing, etc then who cares. In many cases (scientific papers, legal documents) you already have to write them in a voice that is not your own. Go ahead and write these in AI. But if you're trying to say something more personal then the words should be your own, AI will always try to 'smooth' out your voice, and if you care about it, you gotta write it yourself.
Now, how do you use AI effectively and still retain your voice? Here's one technique that works well: start with a voice memo, just record yourself maybe during a walk, and talk about a subject you want, free form, skip around jump sentences, just get it all out of your brain. Then open up a chat, add the recording or transcript, clearly state your intent in one sentence and ask the AI to consider your thoughts, your intent and ask clarifying questions. Like, what does the AI not understand about how your thoughts support the clearly stated intent of what you want to say. That'll produce a first draft, which will be bad. Then tell the AI all the things that don't make sense to you, that you don't like, just comment on the whole doc, get a second draft. Ask the AI if it has more questions for you, you can use live chat to make this conversation go smoother as well, when the AI is asking you questions, you can talk freely by voice. Repeat this one or two more times, and a much finer draft will take shape that is closer to what you want to say. During this drafting state, the AI will always try to smooth or average out your ideas, so it is important to keep pointing out all the ways in which it is wrong.
This process will help you with all the thinking involved being more up-front. Once you're read and critiqued several drafts, all your ideas will be much more clear. Then, sit down and write your own words from scratch, they will come much easier after all your thoughts have been exercised during the drafting process.
No more LLM voice! Every now and then you might get a nonsensical artifact or two, but that's the price you pay for better groceries.
Technical skills aren't worth shit and never were, except as far as they enable you to engage with other people in ways you and they find valuable. This used to be really far! It made it easy to believe that the technical skills on their own were what was valuable.
But the engagement was always the valuable thing, generating technical insight was a necessary prerequisite for it. Now anyone with a $20 claude sub can generate technical insight into niche problems, so what's left is everything else. Everything of value.
I think you're misunderstanding that anxiety. I think this is what's going on:
1. For a lot of people, using technical skills like coding is the part of the job they enjoyed.
2. Closely related to the above point: for a lot of people, technical skills was the part they were good at.
> Technical skills aren't worth shit and never were, except as far as they enable you to engage with other people in ways you and they find valuable. This used to be really far! It made it easy to believe that the technical skills on their own were what was valuable.
> But the engagement was always the valuable thing, generating technical insight was a necessary prerequisite for it. Now anyone with a $20 claude sub can generate technical insight into niche problems, so what's left is everything else. Everything of value.
3. More broadly: if "they" don't find the "technical skills" you provided valuable anymore, why do they need to keep you around? Why not cut your 10-person team down to 3 people (the manager, architect, and technical lead) and fire you and the other 6? Not everyone can re-invent themselves as a product person, even if they had the skills. It's a game of musical chairs.
It really is ultimately a labor relations problem. A lot of people thought they were guaranteed to not face penury and now they're realizing they may have been wrong about that. The issue is that it's ultimately not under your control whether you can sell your labor, and there's no universal fallback if you can't. Solidarity and concrete power for workers are the answer to this and I don't think there's another one.
For the rest, yes, certainly. The vast majority of people do not enjoy their work, are not valued or rewarded for the skills they apply to it. Again, if you thought your job was writing code this sucks. But you're merely one of the billions now. I understand the grief for what people are losing but very very few people ever had that rare privilege.
The problem with LLM voice is that it sends the signal that it's a likely waste of effort to try to figure out if there's any "technical insight" there, because more often than not, there isn't.
Maybe it's not fair that you're paying the price for others using LLMs to abuse their readers, but it's the reality that you have to deal with. It's also not fair to expect readers to wade through all that abusive LLM prose in order to find your (assumed) non-abusive post.
Once LLM writing becomes indistinguishable from human writing, it will be really sad. There will just be a collapse of human engagement, and you might as well just talk to your machine because then at least you know what you're getting and aren't being deceived.
An LLM prose article ABSOLUTELY can mean that. To attempt to conflate the two things is maliciously misleading.
I can’t distinguish between well meaning slop and blogspam slop. I might be able to if I’m well informed on the topic and can point out inaccuracies. Otherwise I just don’t know what is accurate and what is filler or outright wrong. LLM-isms are very distracting and exhausting to sort through. It just doesn’t make for a good reading experience.
There are few things (relatively speaking, I mean online activities, etc) more annoying than seeing a click-baity headline, and then having to read 17 paragraphs to actually get to the point that I realize I already knew and understood it, and it's just slop padding.
It's the modern equivalent of going to a thesaurus and replacing all your natural words with five-dollar ones.
If you're not confident in your English then just write your brilliant ideas in broken English. Or at most do a direct translation. But don't fluff it with AI's nonsense.
If the ideas are good they will be recognized.