## Highlights Their ideal situation isn't one where you visit distinct websites with content created by human beings, but a return to the dark ages of the internet where most traffic ran through a series of heavily-curated portals operated by a few select companies, with results generated based on datasets that are [increasingly poisoned by generative content](https://futurism.com/ai-trained-ai-generated-data-interview?ref=wheresyoured.at) built to fill space rather than be consumed by a customer. The algorithms are easily-tricked, and the tools used to trick them are becoming easier to use and scale. And it's slowly killing the internet. ([View Highlight](https://read.readwise.io/read/01hteqb6eh9bsc8k9d70fx6ve5)) - 💭 thats why there is extreme value in community-driven outlets, in digital gardens, authentic creators, etc This is a phenomenon that Jathan Sadowski calls "[Habsburg AI](https://twitter.com/jathansadowski/status/1625245803211272194?lang=en&ref=wheresyoured.at)," where "a system that is so heavily trained on the outputs of other generative AIs that it becomes an inbred mutant, likely with exaggerated, grotesque features." In reality, a Habsburg AI will be one that is increasingly more *generic* and *empty*, normalized into a slop of anodyne business-speak as its models are trained on increasingly-identical content. ([View Highlight](https://read.readwise.io/read/01hteqcjkfsnpenqma7aqb1z5r)) - 💭 would it be inefficient to continue training models on datasets then if a lot of the internet is starting to be AI-generated? There are simply too many users, too many websites and too many content providers to manually organize and curate the contents of the internet, making algorithms necessary for platforms to provide a service. Generative AI is a perfect tool for soullessly churning out content to match a particular set of instructions — such as those that an algorithm follows — and while an algorithm can theoretically be tuned to evaluate content as "human," so can scaled content be tweaked to make it *seem* more human. ([View Highlight](https://read.readwise.io/read/01hteqmmpd595kz7q51fdgy8zx)) ^c53nbg Amazon doesn't have any incentive to get rid of low-quality books that sell for [the same reason that it doesn't get rid of its other low-quality items](https://nymag.com/intelligencer/2023/01/why-does-it-feel-like-amazon-is-making-itself-worse.html?ref=wheresyoured.at). People aren't looking for the best, they're looking to fulfill a need, even if that need is fulfilled with poorly-constructed crap. ([View Highlight](https://read.readwise.io/read/01hteqqdzkfhnb92hdens3a6ng)) We're watching the joint hyper-scaling and hyper-normalization of the internet, where all popular content begins to look the same to appeal to algorithms run by companies obsessed with growth. ([View Highlight](https://read.readwise.io/read/01hteqs65640rrjd0e8t9bcpf7)) This isn't a situation where these automated tools are giving life to new forms of art or interesting new concepts, but regurgitations of an increasingly less unique internet, *because these models are trained on data drawn from the internet.* ([View Highlight](https://read.readwise.io/read/01hteqsp4bq71t61dfge0xtj01)) [Amazon now summarizes reviews using generative AI](https://www.theverge.com/2023/8/14/23831391/amazon-review-summaries-generative-ai?ref=wheresyoured.at), legitimizing [the thousands of faked and paid-for reviews on the platform](https://www.wired.com/story/fake-amazon-reviews-underground-market/?ref=wheresyoured.at) and presenting them as verified and trusted information from Amazon itself. ([View Highlight](https://read.readwise.io/read/01hteqth6kxyk2s0c6cx2gr0bs)) video may continue to dominate the internet as text-based content finds itself crowded out by AI-generated content ([View Highlight](https://read.readwise.io/read/01hteqxxh3f6mp7mhj830f6r14)) Yet the world I fear is one where these people are allowed to run rampant, turning unique content into food for an ugly, inbred monster of an internet, one that turns everybody's information sources into semi-personalized versions of the same content. These people have names — Sam Altman of OpenAI, Sundar Pichai of Google, Mark Zuckerberg of Meta (which has its own model called LLaMA), Dario Amodei of Anthropic, and Satya Nadella of Microsoft — and they are responsible for trying to standardize the internet and turn it into a series of toll roads that all lead to the same place. ([View Highlight](https://read.readwise.io/read/01hteqz8ye7jt2hdazsrz4jrez)) - 💭 i really like the way he worded this. semi-personalized i feel can have benefits if you're trying to make things more intrinsically exciting, but in terms of development, then no Don't accept that AI "might one day" be great. ([View Highlight](https://read.readwise.io/read/01hter1ka4w607x68zt2a0p1h2)) - 💭 pay for what it provides now, only buy into it when it has proven to actually be helpful