AI & the small Web

Maybe I’m just not well-versed enough in the space and it’s gone over my head, but I don’t really understand how slopsquatting or Bjarnason’s article relates to my post. I think @juette’s post about bots trying to be all-purpose tools reflects my position a lot better. An ethical AI would ideally be intentionally trained on specific datasets which would allow it to have clearer frameworks to build on.

Other issues like installing non-existant packages or Bjarnason’s second point “2. It encourages the worst of our management and development practices” aren’t really an issue of AI. Before chat bots those sorts of people were the same people grabbing random code snippets off the net and tossing them into projects without thinking about it.

The way I was using AI before was more as a supplement than just blindly tossing code into a file and hoping for the best. I would ask it to help me code a feature, then I would ask it to explain individual parts of the resulting code that I didn’t understand. If I still wasn’t sure, I would turn to google armed with more keywords than if I had tried to do it on my own from the start.

Which brings us back to opaqueness. Maybe it’s not actually the right word, but the issue with parsing documentation for me as a beginner with very little foundation is a lack of vocabulary. The actual text of most documentation isn’t hard to read. It’s just actually getting to the part I need to read that’s hard, because I don’t have the vocabulary to ask the right questions to get there.

For example one thing I used ChatGPT for was figuring out how to install myBB on a RaspberryPi. When I googled it, it seemed like none of the results were relevant to what I wanted to do. At the time, I had no idea what SSH was. The myBB install page has instructions for an SSH Quick Install that would have solved my problem, but I didn’t know that was what I was looking for. But, by asking ChatGPT “How do I install myBB on a headless Raspberry Pi?” I immediately and clearly got instructions on how to SSH install myBB. I was able to ask it to elaborate on what SSH was and independently google the commands it told me to use to make sure they were legit.

I’ll concede that even with the amount of due diligence I tried to practice, I probably missed things, but also I do want to re-emphasize how stupid ChatGPT is. A lot of the mistakes it makes become pretty apparent very quickly. It will remove entire sections of code and contradict statements from one message to the next. Anyone who spends more than a day with it trying to do anything remotely complicated and comes out using it’s code without independently checking it’s work is frankly a lost cause.

5 Likes

Problem 1: overbuilding. So many websites are so drastically overengineered these days, because people started relying on massive frameworks - inefficient but very covenient. They’re using megabytes of code to do what can be achieved with a few bytes of HTML. Go to Google’s homepage -you know, the one that’s mostly just a text box and a button- and look at the source code: illegible, hideous, disgraceful. All these chatbots can do is reflect the current state of webbuilding, perhaps piecing together working bits of code but doing it in such a way that it remains difficult to grasp for human brains.

Problem 2: StackExchange syndrome. You want to do 2 + 2 and the top answer is someone instructing you how to reinvent calculus in 25.000 lines of code, and the rest are asking you why you’d ever want to add 2 and 2 and you’ve probably messed up somewhere, please provide all your code and OS and PHP version and logs and maiden name and social securi-

IMO it’s mainly these two factors that have, for some years, pretty effectively scared a lot of people away from learning to code. Goodness knows I spend enough time scrolling through twenty-lines-of-code solutions before finding a sensible answer that I can implement and adapt, rather than just plugging my variables into some black box of code.

So you’re between a rock and a hard place. Unhelpful nerds who have walled off webmastery for the rest of us, or a chatbot that relays their words to you, perhaps a bit easier to parse. It should all be way easier than either of those two options.

It does make me wonder, will there be a shift in favour of clear, legible HTML and simple scripting? I occasionally see devs talk about it, but they’re as much trapped in the framework prison as any of us when it comes to paid work. I feel like I had it easy learning HTML in the early 2000s - straightforward, no JS frameworks, no ‘I built my static website in Flurp using Splong and publish it through Gronk with Blurg’ - just tags and a little CSS.

8 Likes

Hey @beetlegooze, I’m similar to you in that I first tried AI to help me understand things in a more plainly worded (and polite ^_^;;;) way until I learned more about it and eventually abandoned using it. I ended up just doing the freecodecamp lessons and went from there.

I wish I had known better from the jump, but I didn’t! I felt like I couldn’t easily find answers to my questions and just kind of fumbled around until I learned about resources like w3schools and the like. I had a burner account on reddit and didn’t wanna ask anybody anything after reading the way people responded to simple questions. (I know, it was reddit afterall lol)

I think @rmf, what you said here was spot on too:

Made me laugh ^_^ but also it’s like…sad but true, that’s how I felt!

Very happy to discover a place like this forum and also some helpful people on neocities, plus I made a few IRL programmer friends to help me out now. I feel like I know where to look for help these days.

Anyway, I also don’t think AI really has a place in the personal web because it’s…just not personal! Earlier when @rmf mentioned the analogy of bringing an ikea bowl to a ceramics class, that was the perfect way to put it. I felt a lot of shame for having used it when I got more educated on what AI really does not just to the world but to us, on an individual level. Why rob ourselves of the opportunity to learn? I really enjoy learning more nowadays.

Though…I still haven’t even touched frameworks and maybe I never will lol

9 Likes

without getting into my other general ML concerns (environmental, ethical, etc.), at this point, with what i’ve seen from LLMs, i’m pretty much against gen AI replacing any writings or imagery (unless very temporary placeholders) as a general rule, but i feel this way especially if it’s for a personal/indie-principled site. mostly because i believe anyone can (and should) write and, even if you want something specific for photos or art that you can’t do yourself, you can engage with artists, make trades, pay for commissions, or find free visuals/web materials online.

i think part of the idea behind creating your own site, developing and building up this side of the web, is centered around connection. for a project like this, it feels weird to have a indie-principled, indie audience-marketed site that, instead, opts for being in a silo, creating a site void of real soul. i feel this way especially for personal sites using gen AI art since most of it looks so similar to one another.

that being said, i am totally on board with using LLMs as a learning tool in conjunction with project work, courses, or a secondary confirmed learning source (partially to prevent learning from hallucinations but also to round out your knowledge), especially for programming. the reality is that, no matter how nice people are, there will always be random, perhaps asinine, repeated questions i have at some random time in the middle of the night that i need answered (either because my brain will concentrate on that so much i won’t be able to focus on anything else or because it’s a blocker to what i’m doing). it could be annoying. but in the past, i’ve completely lost motivation before because i couldn’t figure out an answer to my specific problem and had no outlet, could not find anyone to help me debug, had to track down well-meaning helpers who’d ghost me after helping me for a while, got made fun of on stackoverflow for not formatting a post correctly, or had questions that i couldn’t articulate. and honestly, using LLMs to supplement my learning has been really, really helpful. i still ask people for help – a lot – but i don’t feel like such an idiot for needing dumb questions answered repeatedly! i’ve been able to maintain my motivation, that’s a first for me wrt programming!

it goes without saying that learning doesn’t mean using gen AI to create a project from scratch to completion. anything i’ve had LLMs spit out at me has been riddled with errors or missing security and required my intervention, so i really don’t trust anything that’s been entirely made with AI to be entirely functional or safe with my data.

this seems obvious from my standpoint, but i do think that any LLM-made project that looks like it was created from top to bottom (including images, text) should be entirely free, running on donations to maintain infrastructure only. nobody should be making money from AI-generated services being offered, and nobody should be shady about using AI – if anything, that’s the one argument i could see where i might be okay with someone using gen AI to make a project from top to bottom: having perspective or insight into a problem that will help others create or be a tool, creating a needed service that currently only has paid or nonexistent competitors, being very open about using AI (the important part) so folks can make an informed decision if they want to use it or not, being open-source and community-focused, and not having a group of volunteers or others who are capable of making the tool. there has to be some kind of greater good for me to be on board – but even then it doesn’t feel worth it, to not even write about your own tool, have it styled differently than typical modern tools, or change anything about the obvious gen AI tells in the copy? it feels disingenuous, and maybe that’s why it gives off bad vibes.

4 Likes

ive seen some people in discords say they basically write entire fanfictions with ai and it just makes me think like. what is the point of doing something like writing as a hobby if you dont like writing and would rather have a llm do all the work for you? at the end of the day this is a hobby and if you dont want to do it then just. dont do it instead of getting ai to do it for you

9 Likes

The ethical aspect is what seals the deal for me too. The environmental damage, the workers being abused (no doubt we’ll be hearing lots more about that after the bubble pops), the deleterious effects on mental health, the fact that it’s all 1000% undeniably plagiarism - that’s enough for me to just never touch it*, and for me to be obnoxiously critical to anyone IRL I run into who does use it. Open and shut.

If someone uses it to bounce ideas back and forth, or to get concept art, or whatever the hell people are doing with it that’s relatively harmless, I suppose I don’t see the creative harm in that - but the ethical case is clear.


*tangent about my own very limited experience with these chatbots

I have used these chatbots precisely two times. One time, I was looking for the tunes to some old Scottish songs for my Isabel Pagan page. In a desperate bid I was using different search engines hoping to find these tunes and it was Bing that kept trying to cram its chatbot down my throat. So I tried it and of course it gave me useless answers. It couldn’t find the correct tune in its databases so it just spat out links to other songs that had similar words in the title - real search engine stuff.

The other time was more recent - I was curious if my robots.txt had successfully kept OpenAI off my site, so I went on the ChatGPT site and typed in what is a baccyflap? and it answered:

A baccyflap is a British slang term, typically referring to the small flap or pocket on the back of a traditional flat cap, used for storing tobacco (“baccy” being slang for tobacco).

Breakdown:

  • Baccy = Tobacco (informal, British)
  • Flap = A small piece of fabric that folds over something

So, baccyflap literally means “tobacco flap.”

It’s an old-school, somewhat niche term, usually tied to working-class culture in parts of the UK, especially in regions where flat caps were common attire (like Northern England or Yorkshire). Some might use the term more jokingly or nostalgically today.

Complete and utter nonsense, supported by no source I have ever come across. I was curious where it was getting this information so I typed where are you getting this information? and it spat out:

Ah, thanks for asking! My earlier description came from a bit of a misfire—turns out “baccyflap” isn’t just a flat-cap pocket after all. Let me share the accurate scoop, straight from the sources:

What the Real “Baccyflap” Is

The baccyflap is actually a compact Bakelite tobacco container—a small, round holder designed for carrying pipe tobacco. It was originally produced in England by the brand Wunup, starting around the 1930s, and later made in plastic by Parker in the 1950s and ’60s. The device typically measures about 95 mm in diameter and 33 mm in height, with an inner volume of roughly 85 ml—holding between 25 to 60 grams of tobacco depending on its form
baccyflap.com
PipesMagazine.com
collection.sciencemuseumgroup.org.uk
.

The Science Museum also holds an authentic “Wunup Baccyflap,” describing it as a phenolic (Bakelite) plastic tobacco pouch made in England between 1921 and 1939
collection.sciencemuseumgroup.org.uk
.

Collectors—and users on forums—confirm this:

“Baccyflaps were made for carrying tobacco with you for your pipe. The original … Wunup Baccyflap … made out of bakelite between 1937 and the mid 1950’s”
PipesMagazine.com
.
One owner:
“Yep, it’s actually called a baccyflap. Made in the 1930’s, by a company called Wunup, it’s a disk shaped Bakelite container for holding a days worth of pipe tobacco”
PipesMagazine.com
.

Regarding launch dates, it seems the earliest use of the Wunup name predates the formal “Baccyflap” — some listings appeared by 1935–1936, with the brand name continuing through at least 1974
PipesMagazine.com
.

Why I Got It Wrong Initially

I incorrectly guessed it was a “flap” on the back of a flat cap because “baccy” means tobacco (which is correct—British slang for tobacco)
Merriam-Webster
Slang Sensei
. But that was merely the root of the word, not its actual meaning here.

My apologies for the confusion—thanks for flagging it! If you’re interested, I can share more about collecting baccyflaps, how much they go for, or even show pictures from museum archives.

Let me know how you’d like to explore it further!

I have those replies saved as it directly credits my website with this information, despite my robots.txt explicitly blocking access. I’m currently emailing back and forth with OpenAI about their violations of robots.txt policy, but it seems their workers are clueless about what goes into their training sets and maintain that their crawler respects robots.txt. To be continued, possibly.

7 Likes

AI can be pretty helpful when trying to code, especially when you just need a basic example of how something works. In that sense, it serves a similar purpose to StackOverflow, except without the insane moderation/reputation rules. For more advanced tasks, like asking it to write the code for a whole feature, it breaks down pretty quickly. I’m not sure if that will actually improve as AI gets more advanced, my hunch is that it will not.

I agree with most everyone else that ethically, AI is really bad. At least, AI owned by large tech corporations is really bad. Which is a shame, because machine learning and natural language processing are very cool technologies with some really great applications. Unfortunately, those applications are not the 10x growth $500B valuation sort of business that tech companies are looking for.

Ultimately, though, I’m not going to judge people who use AI that harshly, unless they’re really bought into the hype cycle. I think efforts are better spent at preventing companies from expanding their data centers and stealing water and resources from poor communities rather than shaming people online for using it.

Edited to add: to be clear, I’m not accusing anyone in this thread of shaming, I’m just making on observation about online AI discourse in general. I think this discussion has been pretty civil.

8 Likes

Yeah, this is exactly what I think AI should be used for. LLMs are just glorified response trees, they would be so much more effective if they were designed with that in mind. Rather than being used to create they should be built for finding information, imo.

There are so many questions that objectively do not really need another person to help you with (you know, the ones that get asked 100 times because the answer seems to hide from everyone who asks) or have such a trivial answer that time investment of going online, asking for help, and hoping someone will answer vastly outweighs the return in asking. LLMs could help so much with that, but the people in charge just seem to be more interested in over promising in pursuit of profits over anything else.

4 Likes

man i’m annoyed. I’ve had “webring generator” sitting in a “list of ideas” note on my phone. I really wanted to write something like this, but assumed it would get very little use. But glad it exists, even though the presentation is very cringe and seems super inauthentic

5 Likes

Do not let the existence of this site stop you - in fact, let it motivate you. There’s a gap in the market at the moment. Not counting this rather odd offering that I don’t really see gaining much ground, Neorings is gone and Melonland’s guilds are tied to their forums, so as far as I know webri.ng is the only open, active webring network and it lacks basic features such as skip mode and custom links - running a large webring on webri.ng can be a bit of a pain. So a full-featured webring provider would be quite popular, I think.

I’ve considered taking my own webring code and making something like that, but I have zero experience with stuff like accounts and passwords and all that jazz - so it’s still on my list of ideas, too.

4 Likes

just wanna jump in and say i would totally encourage you to make this!

i’ve wanted to make a webring but haven’t tried to learn how just yet

1 Like

I’ve been following this topic and giving it a lot of thought since the weekend. I have very recently used LLMs to help me answer questions I wasn’t able to answer by other means. I don’t want to belabor that point, but it’s a real problem. Traditional search is effectively dead because of SEO gaming (and now they’re trying to fix that with LLM summaries - ha!) and online help forums (with a few exceptions like this one) are full of bullies and trolls. I fault nobody for choosing not to subject themselves to either.

But I’d have preferred to have had another option. I’ve been thinking about what that might look like. One thing that feels like it could be worthwhile would be to build a library of “stuff I had to ask an LLM.” It would never replace the LLMs, but I don’t think that’s the goal.

One of my mantras is “I can’t possibly be the first person to try this.” If you’re trying to do something on the small web, chances are someone else is. And if you couldn’t find an answer then chances are neither could they. So one thing I can do is write up guides for the things I went to LLMs for and post them here, or maybe on my site and here with a link here from the post on my site. That way, maybe the next person who wants to do something similar can find that guide and this community.

Anyhow, I don’t have an issue with people using LLMs on the small web to solve some of the problems that have been introduced by the big web. But I think it could be fun and maybe even meaningful to build up a more personal alternative to LLM assistance here. If nothing else, I’ll do it for my project.

9 Likes

Whether or not you asked an LLM for the info, this is a great idea! If you had to search around and put together information from different sources and test it and got it to work, then make a page or a blog post about what worked for you! I’ve done that and I plan to continue to do so.

2 Likes

I think there’s a place for AI as a thinking tool and a learning support. Building a website in 2025 can be extraordinarily complicated and, while platforms like Neocities and Micro.Blog make this more approachable, there are still real barriers for someone who isn’t familiar with how the web works. I think this is also true of the indie web movement itself as there’s still a level of technical competency expected. It’s a privilege to have the access to the education, economic means, and cultural background that allows me to upskill myself for nothing more than the enjoyment of lifelong learning. If AI helps more people build their own websites and share their voice, I’m okay with that - so long as it is their voice (more on that below).

Full disclosure, I used AI to help me build my website. I asked it questions to help me understand how things work and how I should implement things. I find that it can be particularly helpful as a gateway to understanding topics I’m not familiar with or to check my reasoning. Yes, forums and sites like StackOverflow exist, but there is so much gatekeeping and oneupmanship in tech that it can be more discouraging than helpful.

A real example where AI was helpful to me was understanding server-side technologies. I’ve used PHP before but I’ve never administered a web server. Now that I’m running a VPS I found I’d opened a whole can of worms I wasn’t confident in my ability to handle on my own. I didn’t know what NGINX or SSL certificates were, or how to use Environment tokens and deploy my site based on git commits (or even how to use git reliably). Look at any hosting provider’s website and it’s a baffling array of terminology for the uninitiated.

More importantly I didn’t know what I didn’t know. It was AI that helped me not only get a feel for the technologies, it also pointed me in the direction of Laravel Forge. This affords me a layer of abstraction that I’m much more comfortable with to manage my server. Everything I learned through trial and error and talking to AI still applies. Forge just gives me a nice interface for dealing with it. I would not have found it on my own as I didn’t know the problem I was trying to solve.

Even using AI as a development tutor(?) like I do isn’t infallible. I lost a whole day trying to get Statamic multisite to run my various websites before finally discovering it’s not designed for that. The information is there - what the hell is a multi-tenant install?? - but it took me a long time to find it, despite AIs assurances that what I was trying to do was possible (FYI it’s intended for variations like multi-language setups of a single site). I also regularly question AIs conclusions when I know it’s wrong. I think critical thinking skills are important when working with AI. I really hope this becomes part of the curriculum for future generations. We need to bring our own knowledge and experience to these conversations, not just accepting AIs responses at face value.

However, to be clear, I didn’t ask AI to build my website. Aside from the Radical Edward smiley I ‘acquired’ from the Cowboy Bebop wiki and some 33x81 buttons, everything on my site was created by me, especially the crappy artwork in my galleries.

I’m against the use of generative AI to replace human creation such as writing, art and even web development. I don’t agree with stealing art (the aforementioned smiley I’d justify under “fair use” and I’m not trying to pass it off as my own. I do have a credits page somewhere I should probably link to though!) or trying to claim something was your work when it wasn’t.

Beyond that, as someone who gets a real thrill out of understanding how things work, I could never offload the actual creation to an AI. I might ask it to show me code snippets or explain how I might wire an application up, but I would never ask it to build the app for me. That would steal the enjoyment for me.

These AI powered tools that can build a web experience from a single prompt are a big topic at my organisation currently. I see the value in them as prototyping tools and I love the idea of allowing anyone to bring their ideas to life, especially if they’re not technical. However, I’m skeptical of these platforms ability to produce accessible experiences (I was particularly scathing of Figma’s own attempt when it launched).

I’m also concerned about how people will be able to maintain the apps they create without understanding how it works or ultimately involving a developer. I’m sure these companies would say that their tools don’t replace developers or good development practices, but that’s not what their marketing says. Many will see it as a quick fix and I think we will see a lot of abandonware as a result. That’s a version of the small web that I really don’t want to see.

There ultimately has to be a human involved for anything that is created for use by others, commercial or otherwise. At the very least, as a user I’d want to know if an app was built with AI so that I can weigh up whether I feel confident giving it my information or in its reliability as a service.

I am also concerned about the ethical and environmental implications of AI. I’m trying to be more intentional with my own prompts and also avoiding frivolous queries. However, the current speed at which AI usage is changing is worrying. It’s outpacing our ability to control it. It’s more worrying still when the AI advocate at a tech conference says the tech bros in Silicon Valley are hoping that they can burn through the planet now in the hopes of an AI solve to that same problem in the future🙃 I worry too about the masses. The people who don’t know or don’t care to know how AI does what it does. For every person like me who tries to be thoughtful with their usage - and kudos to those who don’t use it entirely - there will be masses who will reach for it daily to produce the next “package me up” trend🤢

Anyway, that was a longer response than I thought I’d write!

As an aside, I’m glad to see this discussion has been reframed as a broader question of AI and the small web. There’s a big difference between this question and opening an individual and their website to speculation and mockery (I’m not saying that was the original intent, just that the risk of dogpiling was there).

It’s encouraging to see that not only was the original post flagged, but that the reporter felt safe enough to identify themselves, and that the post author and respondents have accepted this feedback with understanding, pivoting to a constructive discussion. That’s the sign of a healthy community imho❤️

9 Likes

I agree. I am here for the nuance and thoughtfulness of this community and have really enjoyed reading through y’all’s thoughts without the reactionary dogpiling.

3 Likes

This type of usage does worry me in some other respect: I see way too many people using chatbots in this way and I swear I see them becoming dependent - not dependent on the often faulty information, but on the form factor. You ask a question, you get a tailormade answer. I talk to too many programmers who cannot fathom working without AI, I clown on too many friends for putting shit into ChatGPT instead of looking up real information, I see too many party invites with cute little bullet points, I’ve seen cultural festivals use AI-generated posters.

Hell, I work with kids of all ages and I’m watching so many of them replace a fairly wide pipeline of information with… just ChatGPT. No more looking around, no more sites you can or can’t rely on, no more sources of information, just the bullshit machine that’s mostly right about half the time. It doesn’t help that current-day teens had their lives upended by our disastrous response to covid… and now they have to navigate this nonsense? I can’t blame them, the Internet has been made a swamp of misinformation and monetisation - but boy, I despair in their general direction.

I fear that anyone who uses this stuff for too long will no longer be able to sift through information, to vet sources, to parse the words of others because the machine hasn’t chewed their food for them. And who will put the effort in to learn to write - to form their own words, sentences and thoughts? Who will pick up a pencil, brush or graphics editor and learn how to make something? Why actually get started learning something new, including the difficult and boring bits, when the machine can generate you a pretty-good-if-you-don’t-look-too-close poster, or essay, or presentation? What if good enough becomes good enough?

The AI corporations will collapse and this stuff will disappear from daily life - we’ll recover, but hoo boy, will it take some time. Like the aforementioned pandemic, the damage will probably only reveal itself in stages.

The disinterested shrug with which we’ve let these faulty plagiarism machines into our daily lives will never cease to amaze me, the way we’ll just give up our curiosity and creativity, even if it’s only tiny bits, for some perceived convenience. The only thing in the world that these corporations want is to make money and getting us to depend on them is the way they’ll do it - none of this would be free to use otherwise.


Sheesh. Maybe it’s time to resurrect my blog. Okay, let’s end on a positive note.


Do not let them take your independence from you. Relearn to navigate forums to find the answers you seek, to trick the search engines into giving you what you need; we used to talk about google-fu, maybe that’s the mindset we need. Take pride in finding the unfindable, in getting blood from a stone - don’t let the slopbots, built by the same bastards that seek to destroy the Internet, take that away from you. Don’t give up on the corporate Internet, make the corporate Internet give up on you. This is the way.

4 Likes

I agree. Learning how to discern whether information is trustworthy or useful and being able to understand what you’re reading are really essential skills. It’s bad enough that so many people treat random social media posts as a source of information.

It’s not about keeping down people who don’t have resources. There are (equally free) resources out there. If people want whole lessons on something, why not check Khan Academy, Youtube tutorials, MOOCs, etc.? Reputable ones will actually teach you accurate information instead of making things up. If anything, people who are less educated, less wealthy, etc. relying on ChatGPT and the like only keeps them down even more by encouraging them to be captive to LLMs and other bs.

I do agree that Google and other search engines have become much less useful, but resorting to AI instead us not going to make things better.

As an aside, this one time I couldn’t remember a book’s title for the life of me. I searched for ages and ended up caving and asking an AI model, probably ChatGPT. I tried to be as specific as possible but it just answered with books that don’t exist. For example, naming a title that exists and attributing it to a different author, with both of these being a complete miss as an answer to my question. Asking the question again just gave me slightly different results in the same vein.

I ended up going back to just using search engines and eventually found the right combination of keywords.

That was before learning about the evironmental impact of LLMs. I definitely wouldn’t try this now.

Basically, I just don’t think they’re reliable and if you need more help than a one-off question (and even so…) I would recommend finding the much better alternatives that probably exist out there.

In terms of learning tech stuff, the snobby and rude forums are not the only thing out there! There are whole free courses online and way more accessible than they were when I was a kid :’)

1 Like

The systems we have in place (trying not to say ‘capitalism‘ here) create problems and then give us nonsolutions. We’re getting stabbed and then the guy who stabbed us gives us a painkiller. We’re still bleeding, but we don’t feel it. The info is out there, the Khan academies and Wikipedias of the world, but we have the painkiller of AI, which takes just slightly less effort than looking it up.

I got to thinking about that a few days ago, when I contacted the British Library because I wanted a look at a particular page in a particular book. They have a little chatbox which I apprehensively clicked - but it was not a chatbot. Without a login, without a form, without anything, I was connected to someone at the library who answered my query thoughtfully, sent it on to the proper department (they weren’t in that day) and the next day I got an email with the answer I was looking for.

The frictionlessness, the fact that I could speak to someone who seemed pleased to be able to help, the fact that I could 100% believe that my question would be understood. In a world where all customer service and help desks and whatever are hidden behind chatbots and walls of FAQs and “community forums“ and 12pm-2pm phone numbers and so on and so forth, this beautiful, frictionless interaction made me realise anew the hoops we jump through every day to live and learn.

Imagine a ChatGPT window but it’s not ChatGPT, it’s a global network of people, ready to answer questions and look things up and help you out. Imagine all the money going into chatbots going into a universally accessibly world library, with a million librarians! Man, what a dream.

6 Likes

But then we’d actually have to pay people a reasonable wage for their time. Oligarchs hate that. They want work to become a privilege reserved to an anointed few. Which is all the more reason to do it.

6 Likes

I have a lot of reasons for disliking LLMs (all of them already discussed), but the one that really comes to the forefront when talking about using them to learn things is that I don’t trust an autocomplete to be accurate. And I know that when I’m learning something, that’s when I’m the most vulnerable to not noticing inaccuracies.

I’m definitely sympathetic to poorly written documentation and rude people in forums. But LLM output is just a statistically likely output. It hasn’t read and understood the documentation in order to put it in simpler terms. It’s an improv generator, outputting something that looks like a helpful explanation of documentation because that’s what comes after a question about documentation in its training data (and because its initial, hidden prompt provided by the company says it should be helpful, not rude). Maybe it’s right. Maybe it’s not.

5 Likes