Isn’t it also destroying the internet with low-quality content and affecting content creation in general? Can LLMs still rely on data from the open internet for training?
I'm going to take issue with AI destroying the internet. Our short attention span profit driven culture was already well on it's way to trashing everything that was good. AI is only accelerating the inevitable.
> that's like saying we were going 10kmh, it's nbd that we accelerate to 1000kmh since we were gonna hit the wall anyways
Devil's advocate: folks will take that wall a lot more seriously at 1,000 km/h.
"At Jena and Auerstedt the backwardness of the Prussian Army became apparent. By 1806, Prussian military doctrines have been unchanged for more than 50 years—tactics were monotonous, and the wagon system was obsolete" [1]. They had been obsolete for some time. But they didn't break until they hit Napoleon's army.
Similarly, we have a lot of social plumbing that became–with the benefit of hindsight–obsolete with social media. It was possible to ignore, however, because the rate of change was slow. Now it isn't.
social media in uk law = any website with a message feature.
social media in common parlance = anything from whatsapp and discord to youtube and tiktok, even sometimes aliexpress. anything with a doomscroll feature. anything showing videos, messaging between users allowed or not.
its common usage is confused, the same as the common usage of trolling has strayed from its original meaning because normies picked it up and started using it to mean anyone abusing someone from an online account.
> because weve been able to send messages to each other on a computer since the 50s...
My first partner was born in the 70s and didn't even have a landline growing up.
Here's some stuff I think counts as "plumbing" (i.e. infrastructure) of social connections, which has been lost since the 50s:
• Local newspapers (everywhere, I think?), where an actual editor could (and to a limited extent was held responsible if they didn't) filter out the conspiracy theories.
• Village churches (that might be mainly a UK-specific thing, IDK?) and other similar local community groups, where your local decisions couldn't be brigaded and overwhelmed by fans of a billionaire living on a yacht, as those fans would need to travel to your village personally and most people couldn't be bothered. Now, even when the groups still exist and meet, they can be brigaded.
• Yellow pages getting replaced with Facebook et al insisting that their ad system is the only way any small business could possibly get their name out, when a significant fraction of ads are outright scams.
i feel like peter hitchens saying this, but i agree the british social life is decaying. i thought you meant something else by social plumbing. i cant argue other than i dont think the village churches actually do prevent brigading, locally anyway, and its that small minded middle-englander mindset that the internet actually smashes. my nan was all bout dat life, and they ostracized hetero divorcees, and you can imagine what they thought of gay people... so if you agree with them its fine, but if its your only choice of social group its not so great. ill take the point there is very little public space where people can even assemble to create a group in the first place though
This is exactly how we collectively "solve" so many problems today though, its far from unique to this topic.
We over medicate people, especially the elderly, because each new med has side effects and they're dying eventually anyway. We print more and more debt to paper over massive budget surpluses because the unspoken reality is that we're financially screwed either way. We pile more and more regulations on because we'd rather further grow the government and kick the can a few more times. We bolt one new emissions system after another on our diesel engines because they're already unreliable, who cares.
We don't consider how we got here, only what the next step we take should be. And don't even ask where a step should be taken, progress requires changing things constantly and we rarely give ourselves time to look back and retrace our steps.
Your examples are not supporting your premise. Over medication is from all the attempts to fix all the various medical conditions found. Adding regulations are to fix all the problems of people finding new ways to abuse the system.
This is entirely opposite from accelerationism, which would advocate for less medication so that sick people die quicker, and less regulation so that society would be exploited faster and collapse faster.
Well, then our disagreement is that I feel we were already going at 1000km/h. Nowhere did I say we should keep doing this or it was a good thing or we should ignore it. My point is simple: we already needed stop a long time ago.
Let me re-use your analogy. We were already driving off a cliff, and we are trying to blame the fact that we're pushing on the gas and accelerating however we're ignoring that we were already heading that way and brake lines were cut.
Beat me to it. Facebook/Meta, Twitter/X, Google/YouTube, and TikTok have done quite a bit more damage to the Internet than AI.
The future of the net was closed gated communities long before AI came along. At worst it’s maybe the last nail in the coffin. But the coffin lid was already on and the man inside was already dead.
AI is, I think, more mixed. It is creating more spam and noise, but AI itself is also fascinating to play with. It’s a genuine innovation and playing with it sometimes makes me feel the way I did first exploring the web.
I can download uncensored models pretty easily. There’s even uncensored frontier models. My machine isn’t big enough to run those but you can rent power to run them pretty cheap if you want.
Are you kidding? Look at who runs and funds OSI. It's a revolving door. The main purpose of OSS for the last 20 years has been to "commoditize your compliments" and/or dump on the market to destroy competitors. Any license that attempts to restrict this behavior and prevent billion-dollar companies from simply strip mining OSS is "not OSI compliant."
The entire OSS "cloud native" ecosystem is an on-ramp to expensive managed cloud. It's intentionally designed to be complicated and arcane to sell managed services. Sure you "can" run it yourself, but wouldn't you rather Google or Amazon run it for you?
The main role of OSS in the ecosystem is as a parts yard to support SaaS. SaaS is the most closed model of software development and sales, far more closed than closed-source commercial software you run on your own system.
The OSS mindset is generally stuck in the 1990s and has not updated its understanding of the world since then.
Agreed: The Internet has long been up-to-your-eyeballs with low quality content (i.e., bullsh-t). Blaming LLM software for it is ignoring the well-known reality of just a year or two ago.
Nope. You just miss the millions of SEO websites that was normally easy to spot and to ignore. Now you have millions AI generated SEO webites that are difficult to spot and only contain slop that doesn't help to find the information you search.
This is the same stupid reasoning that told us Trump would be a good outcome because the system was imperfect and ruining it fully would magically create a better one.
I didn't say this was a good thing, I only said things were already fucked. And Trump is also a symptom of a deeper rot in our system. He just happens to be the asshole who took advantage of it.
If you don't fix the deeper issues, it doesn't matter what's going to happen. Blaming AI is blaming a symptom, not the cause.
Stating that we need to fix the deeper problem isn't even close to the same thing as whatever this nonsense is you responded with.
It doesn't have to be low quality. It really is another tool like any other. You can put low effort in and get working results. This low effort, working result gets shipped immediately and gives the whole process a bad wrap. The source is generated crap that lacks craftsmanship and quality. But this gets AI dismissed when it shouldn't be. You can get quality, well crafted source code if you make that a goal and keep iterating.
You can, but when you go through this effort to bring AI to generate good code, you could just self write it. So there are only two kinds of code that are falling out of AI tools. Boilerplate code and shitty code.
Exactly. There's no benefit to using LLMs as they exist today, because it winds up being the same amount of work (if not more!) to ensure that they are giving you code which actually works. That isn't a useful tool.
The Economics of content platforms already started destroying the internet. A lot of the reason the internet was so good for a long time was faith by creators that good content would win, that turned out to be false.
Actually most of the stuff on the internet I really enjoyed was non profit driven. What really destroyed it imo is the attention seeking attitude that results from earning money with advertisements.
Places like microsoft support community forums predate llm and they are filled with wrong and useless information that drown out useful information due to sheer volume. Same with countless websites that scrape forums and other websites and republish the text. Same with auto generated youtube videos - those existed pre llm.
So what's the alternative? Should we go back to reading encyclopedias from the 2010s? I ask this because the need for information hasn't decreased for human beings, just because the capability to produce slop has suddenly increased.
>> I ask this because the need for information hasn't decreased for human beings, just because the capability to produce slop has suddenly increased.
Isn't that the complaint to which you're responding? the SUPPLY side of the equation is the problem, so reading encyclopedias wouldn't impact that. Funny enough the criticism of Wikipedia was that a bunch of amateurs couldn't beat the quality from a small group of experts curating a controlled collection, and we saw that wasn't true. Maybe AI has pushed this to a new level where we need to tighten access and attention once again?