Why can't you just publish the prompt instead? Do you not see how LLMs subtly alter your original message and erase your voice? They fill gaps that didn't exist, they create syllogisms that make no sense, and the voice is now so ridiculously AIdiosyncratic that it makes my eyes boil!
If you have a message that takes 100 words to say, do not use a LLM to add 400 words to it, this isn't a school assignment! Stretching a spaghetti does not yield more spaghetti, it just makes a mess!
Where is the value the LLM adds? Grammar? Vocabulary? The price you pay is you sound like everyone else and your original message is lost in the noise.
As far as publishing "the prompt" - there's no "the prompt." The draft was put together and expanded over a set of interactions with an LLM and other people over the space of about six hours. "The prompt" would have been about twelve pages long and unreadable. Funny as heck, but unreadable.
(If you're really interested, you can check the logs in the site and find the actual interaction that started the article out. It was a comment from someone else, and it got me thinking.)
Heh. Funny thing: I've been writing online and professionally for literal decades, since around 2002 or so, and the LLMs tend to change my actual writing voice relatively little and usually in positive ways, since they say I meander too much.
Writing it myself would mean doing the research myself. How would I do that? ChatGPT can do it faster at scale. Then the summaries are short enough that cutting any particular part wouldn't make sense. I could re-word it, I guess.
Why don't you write all your assembly code yourself? Why do you use a compiler? Why do you generate images, when you can draw them yourself? You're supposed to add value.
I don't think preparing a list of all the threats, editing it and publishing it for others is a "waste". I'm not publishing random stuff, this is important and in line with what I want people to know.
Some people on HN downvote any criticism of AI, other people complain that things are written by AI. If you're such big fans of AI being used more and more, then accept the consequences!
Do the research yourself, the old fashioned way? Search things, write them down, summarize?
The problem with LLMs outputting English is that they're very good at bullshit and it can be really hard to see through the nonsense. The output can be skewed by the model parameters and this can be really hard to spot.
The compiler analogy doesn't work: compilers are (mostly) deterministic and I can verify their output if I wanted to, just ask the compiler to output assembly.
With code generation I can also more or less instantly see if the code is correct or not, because code has less words than human language. The same would apply to images: it would take even less time to see if a generated image is correct or not. That said, I don't use AI for image generation, since I have no use for it.
You're free to write it using AI, but I'm free not to read it. The fact that it's written by AI is a strong signal that the references can't be trusted anyways.
How can you be certain that the ChatGPT "research" you cite is a faithful representation of facts? How do you know that OpenAI/Anthropic/Google haven't introduced RLHF to subtly steer model output on specific topics to align with their political/economic interests?
I'm seeing increasing numbers of people credulously citing ChatGPT/Claude/Gemini output as ground-truth fact. Many more are increasingly lulled into a false sense of security by the citations models append (to the point of neglecting even a bare-minimum skim of the cited sources, much less critically evaluating/contextualizing the nature of the sources themselves). My fear is that most people are blissfully ignorant at the new paradigms of propaganda that AI could enable; most of us here wouldn't be taken by the "slop" image-gen deepfakes (right now), but can you say the same about a couple of citations taken out of context?
We already know how trivial it is to win over a sizeable chunk of society by introducing red-herrings, misrepresenting statistical data, etc. -- oil companies perfected that art, and now as a result a huge number of voters in the US believe that climate change (doesn't exist|isn't man-made|is unavoidable). And that effort was "fully manual" and carried out without the aid of extensive psychological profiling at the individual level via an ad-surveillance complex. Today, society is almost completely defenseless against the extreme granularity/subtlety of manipulation that ownership of frontier AI models enables, especially when it's armed with even a fraction of the torrent of personal data that's being collected on each of us every day.
The people doing the downvoting are different people from the whiners. I'm one of the whiners.
That's kinda fair, like it's still useful to prepare a list, but it's also like if you didn't go research your information yourself why would I start from a position of charitability when I read it? When I research something with LLMs, I know to double-check everything myself before I use it as a basis for my thought or repeat it to other people. Knowing an article is AI written forces me to doubt every sentence. Or maybe it's worse, I have to assume nobody cared about the sentence. The old format was a guarantee that someone gave enough shits to put the article together. Relevance comes implicitly bundled in each sentence. It's like someone talking to you in public in that there's often a reason to pay attention.
It's not as though that person is going to say something correct, or ethical, but I've had a lifetime of dealing with human kinds of wrongness. When stuff is wrong, I'll know it's wrong because the article is slanted or wrong because the author was lazy etc., which will let me discount it selectively and still get value from it when, e.g., a slanted author contradicts themselves. Reading an LLM article I have no clue whether the person who put it up even read the whole thing, so when I read sentences, I have no guarantee that the sentence communicates something worth paying attention to. I dislike that ambiguity and would prefer to guarantee that the text is slop by asking a bot myself. Then I know its worth upfront. I'd be fine with it if these sites included a direct statement in bold at the top: HEY THIS IS AI SLOP IF YOU DONT WANT THAT LEAVE. Then I know exactly how to parse it.
I spent way too much time on actually building this — with Claude and double checking everything — so an article I publish can be OK to push out. We aren’t building a bridge for thousands of cars here, it’s an article.
A lot of things are automated and 95% of the time they are correct. The key is knowing whether the last mile is worth fixing, if the consequences are minor.
Shimman is wrong.
The goal is much bigger, and almost the opposite of what he thinks. It's trying to solve the problem of people chasing "slivers" of money and selling out, which happened in Web2 and Web3: https://safebots.ai/singularity.pdf
What the startup does is make a verifiably trusted, zero-configuration, turnkey environment for businesses to move their data into and run AI workloads on, without worrying about their data being stolen, or some Agents doing unpredictable things. The environment is super-secured, with no ssh. It's an appliance, with over-the-air M of N updates. Think more "Tesla car" and less "OpenClaw". That's the foundation.
That environment then builds everything around a graph database, for people, organizations, and even code. We have Grokers that can ingest a codebase statically once, and then present the graph databases as a far better "RAG" than cosine similarity and pinecone vector databases.
At its most basic level: Agents can't be trusted. We want predictable Workflows, not agents. They can do 99% of everything Agents can, if done properly, and the remaining 1% are the dangerous parts https://safebots.ai/agents.html
It's a lot of innovations at once, including:
Collaborative Bots that are safer than agents.
Workflows and tools that can read, reason and propose actions.
Policies that must be satisfied before actions can be taken.
Logging of everything. Verifiable security and audits for SOC2 compliance etc. etc.
Everything is configurable and designed for serious businesses, not a grandma that finished a Chinese course on how to install OpenClaw on her terminal and not get pwned
> Writing it myself would mean doing the research myself. How would I do that?
This is why you should write things yourself. There is no way an AI would write something so insane in response to that question. Since I can now read your true understanding of the world, I know not to waste my time on your ai slop. I have no reason to believe you fact checked the 'research' done using AI if you cant even understand how the research should have been done in the first place. You want to waste the time of others but arent even willing to sacrifice a bit yourself.
A heat pump is not necessarily dug into the earth. Rather, the flow of the heat pump is moving heat (thermal energy) from outdoors to indoors or the other way around in an air conditioner.
Depending on the direction of the coolant flow, you get either a indoor heating or cooling unit. This is best demonstrated by going in front of the outdoor unit of a heat pump, when they are cooling, the outdoor unit generates heat because it's compressing gas, which then is then expanded when it reaches the indoor unit, generating cold. Exactly like a refridgerator.
Depends what you make it from. If you distill eight litres of wine into about a litre brandy without removing methanol, it has the same amount of methanol than eight litres of wine did. Given the average of 150mg/l of methanol in red wine, this puts it to about 1g of methanol in that amount. That is not healthy, but you need to keep in mind ingestion of alcohol slows down the metabolism of methanol through competition and the methanol will be excreted by your kidneys instead of being metabolized.
So, just like you won't go blind from a bottle of brandy, you won't go blind from distilled wine. However, you're likely to have a serious headache the morning after.
It is a good reminder that most of us that work in software engineering will never build anything remarkable and will fade into history. Even if you do brilliant engineering work for a company, said work was commissioned by the company and the intellectual property rights belong to said company. It wasn't yours to begin with.
I am also glad the
commercial niche illustration markets like Magic the Gathering are extremely hostile to AI art, though of course I would think Wizards of the Coast, the company that publishes MTG, probably see artists as a cost. Maybe.
Perhaps in the future artists will be used to train models that can output a certain style of art and the artist will receive royalties based on their influence on the trained model and its popularity.
Makes sense, though, doesn't it? An elephant's trunk is the fusion of its nose and upper lip, wouldn't that be the location where the mystacial vibrissae (whiskers) be located on any other mammal, making these homologous to e.g. cat's whiskers, which are highly sensitive?
sure, elephants exhibit material intelligence (whatever that means), but the individual whiskers? and also i thought cats whiskers could only determine the width rather than the texture of a gap theyre trying to fit through, though maybe cats also "feel" texture through whiskers
I loved There is no Antimemetics division. I haven't read the new updated to the end but the prose and writing is greatly improved. The idea of anomalous anti-memes is scary. I mean, we do have examples of them, somewhat, see Heaven's Gate and the Jonestown massacre, though they're more like "memes" than "antimemes" (we know what the ideas were and they weren't secrets).
I'm a bit disappointed all names are changed in the new edition. I understand that SCP-... had to become U-..., but I've grown attached to the character names, and they're all different!
I read the original version a few years ago and read the new version when it came out, and I thought that the name changes were pretty amusing. qntm kept the story as close to the original as possible while still making it a legally distinct work for copyright purposes. It's like those off-brand Froot Loops called "Fruit Spins" that are juuust different enough to not get into trademark issues. Except in Antimemetics' case, the "knockoff" version was made by the creator of the original, which I think is pretty funny.
reply