This could be the weirdest kind of moat yet. If you crawled all the things and built a model before everything became bot-generated, you can get clean post-2024 human data from the human inputs to your tool. If you haven't, then maybe you're stuck with the 2023-and-earlier crawls, limiting your models' relevance. We've already seen that the feedback loops of training models on model outputs isn't nearly as valuable, and can get wacky fast. It'll be weird to see how that plays out.
I'm wondering now, does the same effect apply to regular HN readers? In the sense that, we're contaminated (for lack of a better word), and are unable to see things out there without having equivalent connections pop into our heads! :)
The analogy I've been using is an ouroboros of bullshit, consuming ai generated bullshit to generate ai bullshit to consume to generate ai bullshit ad infinitum
Forests that never decayed because nothing could break down liglin molecules for millions of years. Many buried underground and I believe turned into coal/oil.
The shadow libraries are the largest collection of human knowledge to date, and completely untainted by AI. Any search engine that crawls and indexes them will have a tenfold increase in quality and be as revolutionary as the invention of the internet. No LLM model needed.
On top of that, there is no incentive for AI generated content to enter the shadow libraries at all.
> On top of that, there is no incentive for AI generated content to enter the shadow libraries at all.
I think you underestimate just how many people/entities/forces that exist that would love to see further decline, division, and discord in the Anglosphere...
Beyond just western destabilization, there are just flat out people who cause issues just because. Not to mention people who are anti-AI are motivated to weaken AI.
There's no reason people wouldn't taint any source of AI-free information if it became clear that is what it was.
Except that human generated doesn't really seem to matter, all that seems to matter is some basic guard rails on the data you choose. Meta has models generating training data then grading it and select the best examples to reincorporate into the training set, and it's improving benchmarks.
The problem with model collapse is reinforcing means at the costs of the edges of your distribution curve, particularly on repeat.
One of the things that is being overlooked is that offsetting the job loss from AI replacing mean work is that there's going to be new markets for edge case creation and curation.
Jackson Pollock and Hunter S Thompson for the AI generation with a primary audience of AI vs humans, sponsored by large tech and data companies like the new Renaissance Vatican.
Another way they can use this is to log the generated text, and when crawling pages if they find text that Chrome didn’t generate, there’s a chance it was a human, or another tool. But I doubt if people have access to this on Chrome they will really use another tool, so Google can probably differentiate between sources.
I can already see the wonderful cyberpunk future, where people writing e-mails use Gmail's AI assistant to add all the polite boilerplate, while the recipients trying to get through their overflowing inbox use the Gmail-integrated AI summarizer to pare it all back down.
Assuming that, like ChatGPT, the model runs on Google's servers doesn't this vastly increase the cost to Google of offering Chrome for free? Now you have to provide AI compute time to every 4chan poster and forum warrior.
The economics of AI still seems nuts to me. Feels like another bait and switch in the making when all these "free" services need to start showing some revenue.
It's a direct evolution of the search paradigm. You go from entering a few keywords roughly related to what you want and then clicking on ads to continue the search, to having a short conversation with the AI honing in precisely what you need and then having the AI complete the transaction or even generate the content for you, optionally with a transaction attached.
The direct interactions with AI increase the fidelity of the customer model of you that Google has and uses to optimize sales to you for it's customers.
Even further, the most common source of inspiration for purchases is the behavior of other people. If the AI can sufficiently emulate humans and ingratiate itself enough to you then it can directly influence your behavior just by suggesting that it would make certain decisions in your place or that others have already.
This is actually not far removed from the existing situation, just the next level of technological capability.
By actually generating responses for you, it starts training you to allow it to make decisions on your behalf. This may readily extend into purchase decisions.
I find it interesting that the edge browser already has this feature. I wonder if chrome feels pressured to have feature parity specifically with AI or if they believe this change will actually improve their usage metrics?
Little keeping up with the joneses moves like these are always great for a bump in the stock price, its not always to shoot for some metric or business profit
In the example screenshot, the assistant takes this input:
> im interested in this place - do you allow dogs?
and writes this output:
> I'm interested in your property. Its exactly what I've been looking for. To make it perfect for me, I just need to know if the unit is pet-friendly. Thank you for your time and consideration. I look forward to your response.
The input is concise and to the point, the latter is infuriatingly verbose and formulaic. But I guess it'll be easy to filter out humans I would actually be willing to communicate with.
She writes like the latter example. I find myself continuously frustrated by people. She loves them. I find that I'm constantly rejected when suggesting things, she isn't.
I was talking to somebody who worked HR at a multi-disciplinary shop, and she said you could always identify the emails coming from programmers
It was a complaint, definitely not a compliment. She said programmers listed things out in bullet points and bluntly to-the-point. She complained they were dry, intimidating, and she hated dealing with them
I still write concisely and with bullet points, when writing to other programmers. But I now expand things when talking to everybody else. And I've found I get better responses
It shouldn't be terribly surprising that humans incorporate signals beyond pure denotational content of message? Text is a pretty low-bandwidth channel, so we infer as much meaning as possible from the bits of information we receive. All the stylistic choices encode additional information about the sender; part of one's job as an effective communicator is evaluating the effect of all those choices and adapting the entire message (not just its content) to convey the intended impression (not just the meaning).
Incidentally, this is why AI-writing isn't necessarily better communication. The robot can help translate intentions into prose, but it can't decide what one should actually intend to say.
This reminds me of Craigslist. When I get a response that’s written in a terse and grammatically incorrect style, I ignore it. Experience tells me these transactions don’t tend to go well.
There's a middle ground which is what a normal person would write:
I'm interested in your property, but I have a dog. Will that be a issue? Thank you!
or
I'm interested in your property, it looks like just what I need. But I need to know if you allow dogs. Thanks!
People are busy. The kind of filler in the AI example shows that you don't value their time more than you value trying to sound sophisticated when making a simple inquiry. But people also don't have time to decipher possibly cryptic text-message-shorthand. Think about your audience, and write accordingly.
We are still working with yes/no questions. While perhaps a landlord may reply with more information, I would phrase them as open-ended: "What can you tell me regarding dog ownership in your community?" That is an invitation to describe pet deposit, size/breed limitations, places to walk them, etc.
It's not only pointlessly verbose, it ruins the intention behind the input! The user wants to know if they allow dogs, not pets. They can get a "yes we allow some pets" response and now they have to start all over to figure out which pets those are, whether dogs are included, etc.
This is a shitload of computational expenditure to make things objectively worse by introducing an entirely new class of problem to the original message. It's literally "I had a problem, so I used AI, and now I have two problems"
When I take the output apart: The first sentence is to the point and short. The second is potentially redundant but might increase the likelihood of a reply. The third one is perhaps a bit over the top and could be merged shorter with the second (e.g., "... looking for, but I was wondering if ..."). Next one is just basic politeness. Last one feels optional but might at the margin increase likelihood/speed of reply.
Not perfect but not bad either (assuming a human reader on the receiving side).
You can fine-tune LLMs in new styles, without even considering all the styles they are already trained on. The formulaic style response is not needed at all.
The formulaic response in the style of Coding Horror:
"Hey there! Your property has piqued my interest—it's what I've been looking for. Just a tiny detail left to seal the deal: Is the unit cool with pets? Thanks a bunch for your time and consideration. Anticipating your swift response!"
Are younger generations, at least in the US, interested that much in negotiating?
I'm kind of in that age gap where the world started converting to barcodes and computer driven prices and at least to me it seems a lot less haggling occurs now. Again, a lot more of our purchases occur with corporate entities where this haggling doesn't occur. Transactions now are more based on smoothness and speed of transaction. You have X for $Y. Here is $Y. Good day.
Huh, maybe this is why big-G hasn't been too concerned about the rise of ChatGPT. As long as they have Chrome, they still have direct access to a huge portion of web users - even if said users have shifted from using their search engine.
LLMs won't destroy human thought since LLMs are an average approximation of human thought. Sure, this might elevate those who are fresh and are just looking for generic copy, though the best writers are secretly just the best thinkers, as writing is a medium to exercise thought.
I'm a bit biased, having built an AI writing tool myself (https://zenfetch.com), though it's for this very reason that we aren't interested in generating new content on your behalf. We simply want to make it easier for you to recall information to augment your work.
Yeah... at this point I just assume that anyone who can't see any negatives to AI has a financial incentive that depends on their not understanding it.
* Is it people who have trouble with reading comprehension, and don't understand that other people can read a lot more into writing than they do?
* People who are insincere?
* People who think corporate-BS language like "for your protection" and "due to unusually high call volumes" is professional- and smart-sounding?
* People who want to create more utter BS filler in the world for some reason. (See SEO, or the eBay seller feature to create bulk of lies like "the total solution for all your computing needs", etc.)
The only scenario I can think of to which I'm sympathetic is non-native speakers who aren't fluent, and who need a translator, or are afraid of politeness faux pas. But even that has pitfalls: a reader with basic reading comprehension is going to infer things about the 'writer' that simply aren't true. For example, a milquetoast LLM like ChatGPT hits some native idioms, and the reader doesn't realize that there's a huge cultural disconnect in awareness and meaning. Even if the text is even superficially saying what the non-fluent person intended (and even that isn't a given, since they're not fluent enough to check).
Both of the texts are suboptimal in different ways. The original text is:
im interested in this place - do you allow dogs?
Some readers will assume the writer is not well educated because "im" should be capitalized and there should be an apostrophe. Other readers will notice the use of a hyphen, which is not very common in written text; it reveals that the writer may actually be educated but writing quickly. A well educated reader will see both of those signals and recognize that this is too little information to reliably judge the writer.
The AI version of the text is overly formal and verbose, making it clear that the writer does not wish to reveal their level of education. I think that's the reason people might be interested in this.
Its interesting when all you seem to factor in is level of education, implying it's the most important metric for selecting tenants. I'd contend its hardly relevant.
My only gut impressions is the first seems rather nonchalant, which is sort of strange when entering a presumably expensive contract. The longer response just feels very boilerplate to the point where I'd question if its not the opening to a blanket scam message.
I guess an important takeaway is that everyone perceives interactions in different ways and that's really why this whole thing is relevant.
I think the apartment writing is symptomatic of the extreme, highly educated, San Francisco bubble many live in.
The people I know in real estate are not reading Hemingway on the weekend and appreciate fine writing. They are blue collar guys with good business sense and good at fixing housing.
The writing style of the tenant matters to them zero.
"You can write a proper restaurant review for Benu or Birdsong along with writing a good inquiry into getting that apartment in Presidio Heights you always wanted"
I think of it more from the writer's perspective. In my experience, a large number of people shy away from writing anything because they feel they cannot write in a way that makes them sound smart. (And if they're not smart, they believe the recipient will not be interested in helping them.) I think there's value in tech companies helping people overcome that fear.
I think you've found a legitimate use for this. (As a stopgap measure, for better education.)
Sadly, I don't think that will be the majority of the use.
Also, if this were the target use case, the use case could be adapted to the larger problem of tutoring/coaching feedback, to help the person learn and improve, not "write my essay for me, I don't much care what it says, just make me look smart".
The AI version of the text has me expecting that the next reply will be an explanation that they are a Nigerian prince and will be paying me twice my asking price but that I will need to cash the first check and send half the money to an escrow holder to prove that I am trustworthy.
Are we leaving the era when adtech and surveillance by Google were things we could look past, because they mitigated it with some older-era good things?
It is true that Bard/ChatGPT is just two clicks away. But never underestimate the power of defaults. This is definitely not a good default for writing anywhere on the web. Google could at least have made this an extension instead.
The internal combustion engine could mark the end of buggy whip manufacturers.
I jest, but only a bit. New inventions can and do wipe out old sectors, but it's hard to tell in advance if you're seeing a real transition or a pointless flash in the pan, and people make mistakes in both directions.
The example is that it can make your writing more long-winded without adding any important details, so that it takes more effort for the person to respond? Why? I'm already overly verbose as it is.
> Could Mark the End of the Human Internet
Man... what does that even really mean? Popping over to ChatGPT to do this kind of shit is already mainstream enough to have been the subject of a South Park episode. There's probably hundreds of similar browser extensions for Chrome alone. I guess this is more convenient, but what problem does it really solve?
Call me crazy but, I somehow imagine this browser feature will not lead to some AI Internet singularity. It's just going to slide the crap-factor up a few more notches than it already is, making the Internet even less enticing to use.
