The horizon is not so far as we can see, but as far as we can imagine

GIGO: The Past of Google Search Is the Future Of LLM AI Models

If you’re old enough, you remember how amazing Google search was when it first came out and for the first few years. Excellent results, right at the top. Nowadays, it’s crap and half the time to find what I want I have to append “Reddit” or search very specific domains. (Reddit is likely to be worthless in a few years due to the IPO.)

Anyway, Google search results became crap for three main reasons, from least to most important:

  1. Worship of the official and the orthodox. Every time I search some medical issue, the top twenty sites tell me the same thing. That didn’t used to be the case, for cancer, for example, the old “cancer tutor” site would be on the first page. Maybe it’s good that the equivalent isn’t any more, but I wanted to read the alternative views as well as the orthodoxy.
  2. Monetization. Prioritizing selling ads over providing the best search results has had the effect one would expect.
  3. Organic link destruction. What made Google so good at the start is that its algo was almost entirely based on the number of incoming links a site had. Since the internet at that point was almost all human created, links were highly curated and genuine: someone had read the site, liked it and taken the time to link to it. Nowadays, most links aren’t organic: they’re SEO crap or advertising or intended to play the search algo, leading to an endless arm race. A link is no longer an endorsement and there’s no easy way around that: nothing can replace a human being reading a site, liking it, and linking to it.

Google, to put it simply, destroyed its own usefulness by destroying the internet ecosystem that had organic links, links by people who didn’t expect to be paid for them, to sites they found interesting whether those sites were official or orthodox or not.

Now, Large Language Model (LLM) AI is based off training on, basically, the entire internet. It’s essentially statistical. How good the AI is based on how good what it trained on is, with a lot of tweaking to try and point it towards stuff that isn’t horrible (not good, horrible, like avoiding racism.)

The problem is that over time more and more of the internet will be AI produces. AI will be feeding on its own output, rather than on organic human writing. It’s already been noticed that AI that eats its own dogfood tends to go nuts, and it’s fairly clear that AI is rather bland: it is a blender for what’s been done before.

So AI will, like Google, damage the very resources it needs to work well, especially since most people won’t admit that their AI content is AI, so it’s hard to avoid. It will slowly degrade over time, with some bumps of improvement as new advances are made.

Mind you LLM AI isn’t a general AI, it’s not smart, it’s just another algo. It doesn’t understand anything. Real AI will wait further advancements, if it every happens at all.

Still, enjoy it now while you can. I expect it’ll get better for two to three years, then degrade. That’s the optimistic view, there’s some evidence that the degradation is already underway.

You get what you support. If you like my writing, please SUBSCRIBE OR DONATE

Previous

Poverty, Wealth and Money In The New Green Age (Principles of the Green Age #1.1)

Next

Update On Server Issues and Load Times

10 Comments

  1. thought dream guillotines

    Thanks Ian.

    I thought’s bruce wilder’s succinct summation last year was instantly clarifying, so I saved the link. This was prompted by Ian’s article from May last year entitled “Google Neural Net “AI” Is About To Destroy Half The Independent Web:

    “AI language models will be cannibalizing their own output very quickly as they flood the same space they gobble up. Because they do not “think” but simply mimic style with no comprehension of substance, a feedback loop that spirals very quickly seems quite likely to me, imitating itself into becoming a formless blob.”

    https://www.ianwelsh.net/google-neural-net-ai-is-about-to-destroy-half-the-independent-web/#comment-146182

  2. efficient idiocy

    Thanks Ian.

    I thought’s bruce wilder’s succinct summation last year was instantly clarifying, so I saved the link. This was prompted by Ian’s article from May last year entitled “Google Neural Net “AI” Is About To Destroy Half The Independent Web:

    “AI language models will be cannibalizing their own output very quickly as they flood the same space they gobble up. Because they do not “think” but simply mimic style with no comprehension of substance, a feedback loop that spirals very quickly seems quite likely to me, imitating itself into becoming a formless blob.”

    https://www.ianwelsh.net/google-neural-net-ai-is-about-to-destroy-half-the-independent-web/#comment-146182

  3. Eric Anderson

    It’s a brave new world Ian, and AI is the new soma.

  4. bruce wilder

    Western philosophy has long wrestled ineffectually with the challenge of fitting form to truth. That form was a fit standard for the validity of logical argument was the great and fatal insight of the Greeks and finding an equivalent standard form for so-called inference became a goal of generations.

    Proving the validity of a logical argument is in some sense, mechanical, because all that is necessary is to show that the requirements of form are satisfied. Computers can be programmed to do it, to find the pattern that is logically valid form.

    I sense that LLM AI is an exercise in finding the “form of truth” where none exists or can exist and what it identifies is who and what apparently sounds authentically authoritative. It is finding semantic forms of truthiness, but is not learning how to distinguish empirical proof from wild speculation soberly expressed.

    Humans, in turn, are learning that they do not have to reason about facts. It is hardly coincidental that this is an era in which disinformation and conspiracy theories pervade the discourse, when Public Relations specialists multiply and journalism is extinguished.

    Semantic pattern recognition is fundamentally misapplied here. It is fortunate that computers are likely to have a tin ear and produce soaring rhetoric that sounds as empty of thought as Obama or a bad listicle.

  5. responseTwo

    Big-money stock market greed ruins everything that was once a superb product. Boeing is a good example.

  6. different clue

    For the longest time I have thought that eventually The Establishment would shut the Internet down for good. Now I think they have found a way to achieve shut-down without actually having to shut anything down at all, or hit any kind of kill switch.

    Just recently at NaCap, Lambert Strether too has started noting that the language-mimic AI algobots don’t know AI output from human output. So as the AI algobots flood the internet with more AI language-mimic output, the AI algobots will be training themselves on more and more AI product. L. S. has started calling it coprophagia.

    So the AI algobots ingest human shinola and excrete AI shit. And when the AIs ingest AI shit, they excrete shitshit. And when they ingest shitshit, they excrete shitshitshit.

    Steve Bannon, who was one of Donald Trump’s disposable advisers , said they way they would deal with the media, which was the Real Opposition, would be to “flood the zone with shit”. So the Big Search companies have created a forcefield-matrix in which the shit rises to the top. And the emerging Big AI companies ( sometimes the same companies) have created language-mimic machines which exponentialize the shit and exponentialize the exponentialized shit and exponentialize the exponentially exponentialized shit, and round and round and so forth and so on.

    They hope to drive natural humans all the way off the Internets through sheer applied disgustogenesis. They hope the natural humans will decide that if the Internet has become a sewage cannon, they ( we) will eventually stop stepping in front of it.

    Since the Lords of AI want to get us natural humans off the internetby “flooding the interzone with shit”, is there a way for natural humans to dodge the sewage and dance between the sewagedrops and keep finding eachother online and share information etc. from human to human? ( Eventually AI will move on to flooding the “image searching” functions with AI images. Natural humans may well be able to train themselves to sense the difference between real and AI by learning to sensitize themselves to whatever Uncanny Valley appearance the AI deepfake images may have.)

  7. bruce wilder

    @ different clue

    off the internet? how about off the planet?

    flooding the political culture — really all culture — with crap, repetitive, mesmerizing but boring, empty calories for the mind

    I like your hope for social connection that somehow escapes the endless flood tide, but how likely is that? The whole point of the “design” would seem to be to destroy any possibility of a powerful community or movement emerging: to detect early and subvert and demolish. And, leave no refuge on the margins. No place to hide.

    It is like someone went to see George Carlin and came away with a blueprint.

    Have you ever read the annual report or website of some do-gooder NGO from the last decade? The mission is collaborating with partners throughout the something-sphere. Really it is beyond parody. And for the Right, that is “the Left”. Empty. Nihilistic much of it. Fake all the way down. And they did that BEFORE AI.

    I think most of us contributed to building it up. I feel guilty about investing in a good opinion of Obama. Lots of people went along with Iraqi involvement in 9/11 and the threat of WMD.

    It has been a long time since the political discourse fed on anything with much reality-content. Maybe since we were asked to believe the Warren Commission or the Gulf of Tonkin Incident.

    But I seem to remember great organizations could do things or support the good work of heroic genius occasionally. Prevent polio. Go to the Moon.

    Competency is gone. Boeing has practically collapsed. The CDC and Public Health is a bad joke. Journalism has putrefied. All before AI has begun to work its magic.

    I guess what I am trying to say is the ability of political society to deliberate together and act in some concept of a general interest has been deteriorating for decades. And that deterioration has been driven in large part — has been composed of — a common willingness to simply not engage seriously with political propositions or proposals, to criticize or think. We just watched it happen on tv. Political commentary became media criticism. In an endless meaningless scroll.

    I do not know if people are genuinely disinterested in intelligence and integrity or if the flood tide of dreck simply takes it out of reach or detection. But I have no hope of many finding and then recognizing it, let alone engaging it usefully.

  8. Willy

    Our problem is the sheer number of people who’ll buy up all the processed shit, oblivious. They’re the target demographic. They’re the domesticated consumerist livestock trained to mindlessly follow the bells and whistles.

    They still pack Disney parks despite costs at 4 times the rate of inflation from when I was a kid, even with all the reports of rampant money extractions and reduced maintenance of the park itself. “Moo” they mumble as they shuffle forward in interminably long lines. Compare pop music of the 60s-80s with today’s bland nonsense. Or TV and movies. Most household appliances. Late-stage capitalism needs those people. So who deprograms them?

    Is it up to the boomers who (as a whole) allowed all this slide into doughy-diabetic fatness? Our woke younger generations? The Joe and Bernie show?

    Me, I still go online to do most of my research. I can still find reasonable answers and innovative amusements. But true, it’s a lot more work for me now. I have to read between a lot more lines than I used to. But doncha hate it when you ask, “So where’s all the good music these days?” and you’re told “It’s still out there, you just gotta look for it.” Maybe it’s up to me to share where all the good stuff is at.

  9. different clue

    @bruce wilder,

    Yes, there are no margins and nowhere to hide from Big Silicon’s digital Feceworld Of Tomorrow without going offline and staying analog.

    That is why I did not suggest trying to hide or find a margin or a shelter. I suggested learning how to dodge the sewage cannons and dance between the fecedrops. Given all the information still findable online, the effort is worthwhile as long as the information remains findable and sharable.

    Going analog means buying and keeping books, magazines, papers, pamphlets . . . and supporting those libraries which still support analog printed matter. Many still do.
    The anti-paper cyberbarians and digigoths have not taken over the leadership of every library yet.

    If there is no hope of many finding intelligence and integrity, there is still a hope of some few finding it as long as it exists to be found and supported. Those few should perhaps view themselves as the latter day Irish Monks, saving the intelligence and integrity parts of the physically reproducible media record by saving it, copying it, storing it, etc. And also finding lo-sewage computer media to stay in touch on and with, and sharing sharing sharing genuine information over those channels as long as they can be kept in existence.

    Perhap a few million shit-free data/info rebels will have, keep and maintain a few million stand-alone computer-set systems and share information by means of tiny little memory sticks tied to the legs of carrier pigeons flying between the memory-stick replicator persons.

    Catapult the samizdata.

  10. capelin

    @ dc

    “Perhap a few million shit-free data/info rebels will have, keep and maintain a few million stand-alone computer-set systems and share information by means of tiny little memory sticks tied to the legs of carrier pigeons flying between the memory-stick replicator persons.”

    Exactly. It’s like an extreme version of turning off “cookies”. Kludges in this spirit are how humans adapt to adversity.

    Less than we think sometimes, but often there’s a bit of autonomous space on the margins of whatever’s going on.

Powered by WordPress & Theme by Anders Norén