Sometime in the last two months I noticed myself going straight to ChatGPT, and Gemini flash when my free credits ran out. And ocasionally some Reddit threads.

Otherwise those countries would be SO fucked! So you can rest assured that their governments have worked tirelessly to prevent that from happening.

I think I'm going to dream of riding flying pink chocolate unicorns tonight.

Maybe it’ll be better this way.

Large amount of content has already moved behind walled gardens. AI is going to kill blogs etc. That doesn't leave much. Bit of niche stuff like lemmy/bluesky/hn I guess?

Both are however in line with my personal experience and observations regarding how people around me are using the web now.

I find myself using web crawling in LLMs a lot more, and search a lot less. My reasoning follows, and I think most people would agree.

- When I’m looking for some relatively obscure information which I’m not sure where to find or which would require hours of research for me to find, I use ChatGPT (usually o3 with deep research) and refer to citations for more information regarding a topic. This saves me hours of investigating, which I usually don’t have for something that’s just a curiosity. A friend also used deep research to find papers highly relevant to a topic he was working on for his med PhD in minutes, claiming that just searching through PubMed to find such papers would take him days - and probably less successfully.

- When I’m looking for something specific in regards to a topic I’m relatively familiar with, I use search (usually Kagi, unless at work where it is banned!) to quickly find reference material.

- LLMs (and engines like Kagi) let you skip through the SEO spam you’d usually skip through when using Google, as well as letting you search more easily (due to better natural language understanding than classical query engines). The quality of search results had been diminishing for years (geeks4geeks ranks higher than SO on Google), so it’s not surprising people turn to tools which produce better results. It’s like being shocked that people are driding cars instead of riding donkeys.

A particular example is that I looked up a DTC that my car was throwing. I googled it first, and got a results page consisting of forums that said nothing, paywalled generated sites that also didn’t provide any info, scammy Scribd clones hosting diagnostic manuals for the wrong car model and ad-ridden garbage sites that just claimed “oh it could be anything, just take it to a mechanic”. ChatGPT gave me an exact (and correct) answer in seconds.

This is an expected result of what we have done to the web, and it should surprise nobody. I’m only sad that genuine, small online communities are dying in favour of walled gardens, but that’s an entirely separate discussion.

I need to switch everything to a font with crossbars on the upper case I.

In this case, the organic content will lose against the invading crawlers, either way. If they build a fort against the bots, they go undergound and remain unseen by their customers. If they don't mind bots, bot summaries would become those forts, with the same effect.

So ultimately the websites have to realize that bots are the only customers they can expect. Humans are long gone, as no human would ever need to click a link, if they are fed with the summary walls.

To realize this, start a blog and see how tough it is to get any meaningful attention to your great content.

essentially whats happening here is that LLMs have removed the NLP barrier to automating information flow. thats what agentic systems are under the hood, once you realize alot of the "agentic" stuff boils rule-based engines and hand built state machines. and it will be until the next AI winter gives us proper dynamic modelling of an envirenment, rather than static next token prediction

i think RAG replacing the traditional reverse indexing will mean information wont be tightly linked to webpages anymore. meaning no ad revenue and SEO dying. which i think is good for the internet, because ad revenue just encourages information rent-seeking. this obviously assumes AI RAG is reliable.

i think without the ad revenue profit motive, information needs to be generated some other way. journalists and bloggers need to be supported somehow. crowdsourcing like patreon and stuff could help, but that just shifts the capitalism from optimizing for sensationalist clicks, to optimizing for cult of personality, which i think will become the norm for information generation unfortunately. everyone will have to make content, rather than news to do this for a living.

but i think that wont have the pull that ad revenue does. so i think information generation shifts alot more to amateur bloggers and taxfunded government mouthpeices

for actual services, i think we get more supercharged APIs like MCPs and that makes economic activity much less frictionless. its hard to know if this will just be a quantitative change or an evolutionary one

ones this is the labour displacement. how would all the people who were doing the jobs caused by the impossibility of NLP automation make a living? i think going back to the idea the "agentic" is just rule based engines with automated NLP, which is similar to coding (which is just making rule based engines with automated Formal Language Processing) i think "prompting" (designing, implementing and maintaining rule based engines in natural language) will become the new coding (which, again, was the same thing but making rule based engines in formal languages). essentially LLMs are just the next evolution of compilers, and people will learn to use them as such.

what i have no idea about is what sort of evolutionary change this will bring up, the way compilers brought when they replaced writing machine code. i think formal language programming will still be done by people, like machine code is still written, but the newly unlocked natural language coders will far outnumber them. so then what will be unlocked by this higher level of programming? like what will be the internet of the LLM era?