Fuck AI

I don't think people realize what's going to quickly happen is that people making the models will start extorting brands to get a better ranking. You want our model to recommend your brand then pay us $X. Then any perceived utility about reading reviews vanishes kind of like with fake reviews today.

hey use me i'm better than google

I don't use copilot (or windows), nor do I believe that LLMs are appropriate for use as a search engine replacement, but to be fair, google is really bad now, and I wouldn't be surprised if people are having a better experience using LLMs than google.

This is how my younger brother already is. Whenever something goes wrong, his first instinct is to open the ChatGPT app and ask it what to do.

Car engine making weird noises? Ask ChatGPT. Botched dinner recipe? Ask ChatGPT how to fix it.

Evidently, he sees nothing wrong with this, and does not consider the fact that the answers regurgitated by the LLM may not even be relevant/correct/up-to-date. I imagine this is how most people today treat LLMs.

This happens because search engines have become worse and worse over the years. Unless you’ve installed adblockers or have a pi-hole. Not to mention that many search results Google returns are just AI generated websites. And the average person isn’t going to pay for Kagi to get a better search engine.

My mother is constantly googling things and reading me the AI overview. And I know LLMs make shit up all the time, and I don't want AI hallucinations to infect my brain and slowly fuck up my worldview. So I always have to drop everything and go confirm the claims from the AI overview. And I've caught plenty of inaccuracies and hallucinations. (One I remember: she googled for when the East Wing of the White House was originally built and the AI overview told her the year of a major renovation, claiming it was the year it was built, but it had been built much earlier.)

I spend significantly more time keeping AI at bay than using it.

• It is not good enough to write code at the expertise level I'm usually required to work at. I've spent more time fixing generated code than it would have taken me to write it from scratch.
• It hallucinates too much for me to trust any information provided by one
• Security of AI companies is about the same as IoT companies with the difference being that if IoT leaks my data it's going to be incompetence and not malice - I don't trust AI with any of my local data.
• AI agents require to be given even more access and permissions and that's just not happening.
• I contact support when I've exhausted what I can do myself and as a result AI chatbots are an annoying obstacle that can't help me and I have to waste my time going through to reach a person that actually has power to help me

Yes. That is how the internet will be used, to a significant degree. Even today AI represents an extremely powerful and astonishingly human-adapted user interface.

The internet brought a vast quantity of knowledge together and made it accessible to anyone, in theory. In practice you need arcane knowledge to get what you want. You need to wiggle the mouse just so, need to know the abstract structure of the internet, the peculiarities of search terms, … it’s eminently doable but it’s not natural or intuitive. You must be taught how to use it.

If you put a medieval Tamil farmer in a room with a ChatGPT audio interface they could use it and have access to all of that internet knowledge.

I understand a lot of the backlash against AI but I don’t get hating on it because of how good of an interface it makes.

I never understood how people order shoes online except if they already have that pair. I go to a real-world shop and try 10 pairs and honestly, 8 of them aren't so great. I wouldn't know how that works with Amazon or an AI.

I have found no use for trivial stuff as I don't trust what it outputs. If I have to verify its answer, I might as well just research it on my own, it's literally more work to ask and research than to just research

At work, the only use I have found is to provide fake data so I can run some tests as I work on confidential stuff I cannot use directly. For example, they other day I asked for 30 super hero name, last name, gender and DOB table just to pick from. I could easily do it by hand as it does not have to be accurate at all, but it was faster to prompt than to just randomly type and I did not care if it missed the prompt (very rare scenario IMO)

Sponsored search results suck, but sponsored LLM results are gonna be wild.

I'm a student, I've strictly removed myself from this fucked up story. I'll be it, in my CS class with about 125 student, I'll say 87℅ of them use chatGPT, probably even more. It's easy to see that almost everyone in the class cheats, I spend about 5 days on a project or lab, they spend only 1 day. Its not that they're good at CS, this is after all an "introductory" CS class, and at the moment were learning data structures. Everyone hates data structures, and the people I talked to said "yea I spent a day on the linked list lab and got 100" it is so rare for me to get a 100 in this class, and I spent 5 days on that lab.

I've completely removed myself from using chatGPT and other LLM's. for fuck sake, I'm using a query based search engine called marginalia search, because even the internet has been fucked with unreliable information.

I love it because now I don't have everything at my fingertips, I do have a few things with the search engine, but you have to conserve the words that you use and not everything will pop up. For example, You can't just search "how to concatinate numbers in c++" you need to say "string concatenation c++" since you're using too many keywords

Because of this, I've started checking out library books from my university, and what's so funny is that the books I checkout aren't due for months in advance because no one checks out books anymore. I got a book on C/C++ reference guide and it isn't due till January of next year.

Search engines have cratered.

There's a need to skim through all the crap to find what's valuable, and AI offers to do that.

I have used it to learn things, specifically how to use Angular. Angular has enough versions, all still used somewhere, that genuinely valid and helpful advice from a few years ago is misleading. AI didn't so much take the place of a tutorial, it took my whole code and reviewed it, telling me what was wrong. Then I fact-checked its answers because I don't trust it, and yeah, it wasn't always correct itself, but it was more than 80% of the time, and even when it was wrong, it got me close enough that I could find the right answer where it failed.

Also, it does a great job with CSS right out of the gate, no mistakes yet on that front.

What? As a person with wide feet, there are sizes that are specifically made for wide feet. This isn't anything they couldn't have learned from a normal search engine.

Someone at work is using Chatgpt to guide her retirement accounts. I just... I'm so tired man. I hope it works out for her but like really?

LLMs are good at templates or starting points for standard documents, or communications and coding examples

BUT ... you have to double check every single word

Ok, but how much are people willing to pay an "AI" to find sneakers?

How long can "AI" grifters dump money/resources into these free services?

If "normal people" had to pay the actual costs, I'm sure that many of them would look for their own sneakers.

Personally I don't need/want a machine to pick my shoes for me in the first place.

Anyway I guess my wider point is that people will stop using "AI" so much when they actually have to pay for it.

I told my students to use it as a help, not for any thinking (teachers have been told to instruct students on using AI). They used it the other day, by taking a photo of a hand drawn table (database) their math teacher hade drawn. AI then made it into a nice spreadsheet for them to use. They did have to account for it not understanding the letters Æ, Ø, Å.

Considering just how many of the top results in any search engine are AI generated SEO farming I can understand why people do it. Finding decent results is HARD these days, especially on subjects you're not already familiar with (because you'll not notice the red flags).

Around me ppl use LLM for pretty much any question they cannot answer, ranging from tech support to life advice.