Addressing the AI situation.
No AI was used to write this document :) I do not consent to use this for training LLMs.
Expenses
I'm nobody to talk harsh about decisions multi-billion dollar companies make. But i believe there's a problem with how this technology is shoved into the throats of customers in a magnificent scale that is it being today.
By AI, I will be referring to the current implementation and delivery of AI by the leading players in the market. Dominated by LLMs and LLM powered Agents.
AI costs a shit ton of compute. In fact the AI powered google search is estimated to cost more than five to ten times the cost normal google search would have(ubs). And additionally, the newer AI overviews feature is supposed to "[use] a “query fan-out” technique, issuing multiple related searches concurrently across subtopics and multiple data sources"(google) - each additional search costing extra compute.
Often the introductory plans - even without considering the free plans - are heavily subsidized by API providers, as the next level of API plans are most commonly way more expensive. It is not new that companies are relying on tokens for measuring and controlling API usage.
Anthropic gives a 5$ free testing credits, which is barely enough to read a considerably small source file. or few minutes or maybe an hour max of agent runtime. In such pricing, usage-based billing can rack up thousands of dollars each month to do anything meaningful. A large source file can end token limits seconds.
Not nice
Step one, make a search engine which indexes websites and puts hyperlinks to them.
Step two, make it better using bots to crawl far more websites and make them searchable.
Step three, add ads and make them appear first.
Step four, scrape all the data from the websites and train an LLM with it.
Step five, move the hyperlinks away and put the LLM first.
Google is killing the web it initially supported, without giving anybody any form of compensation. Reduced traffic for web publishers will make their ad revenue unreliable. What this shows is their absolutely non-existent plans on long term sustainability and lack of ability to understand the fact that you need a constant stream of new data to keep LLMs up to date. But it's not just google, AI chatbots itself have been doing this for a while now.
Increasing amounts of web scraping bots and AI written articles have been intoxicating the internet at an astounding scale. The largest repository of information humanity have ever developed is being made into worthless trash by the same technology which was and is to be trained on it. It is found that AI generated content degrades the output quality of LLMs trained on them (model collapse).
This is like growing a weed that excretes poison in the water it has to drink. It is nonsensical, ironic and real simultaneously. Maybe they are thinking sustainability doesn't matter when an AGI super intelligence pops into existence and decides to solve all the problems in the world. I can't know.
What about the people?
I am not raising the problem of people losing their jobs because of AI. It doesn't make economic sense, we are not there yet. What i find concerning about humans are two different problems, the economic and skill dependence on AI.
By economic dependence, i refer to the massively subsidized and generous free tiers that an incredible number of users depend on. It is simply unsustainable, and if one day the companies became less generous, people who lack the riches to afford AI can seriously get hit hard.
There are counter arguments to this, computing may get cheaper, AI gets more efficient, competition, and alternatives. But it still is a valid concern because the counterarguments depend on things we don't know yet.
Secondly i mentioned about skill dependence, and that is a much worse concern. If people became dependent on AI for writing, debugging, research or memorization, or the worst, critical evaluation: we end up having to rely on companies providing it.
Not everyone can afford to buy the amount of resources AI companies can provide with supply of shareholder money. We became dependent on calculators and it wasn't that much concerning because it was relatively much cheaper. But the economic impact AI can have when people rely on it for everyday decision making can be huge.