• Moderators, please send me a PM if you are unable to access mod permissions. Thanks, Habsy.

Tech Thread

Why A.I. Didn’t Transform Our Lives in 2025​

This was supposed to be the year when autonomous agents took over everyday tasks. The tech industry overpromised and underdelivered.

As 2025 winds down, however, the era of general-purpose A.I. agents has failed to emerge. This fall, Andrej Karpathy, a co-founder of OpenAI, who left the company and started an A.I.-education project, described agents as “cognitively lacking” and said, “It’s just not working.” Gary Marcus, a longtime critic of tech-industry hype, recently wrote on his Substack that “AI Agents have, so far, mostly been a dud.” This gap between prediction and reality matters. Fluent chatbots and reality-bending video generators are impressive, but they cannot, on their own, usher in a world in which machines take over many of our activities. If the major A.I. companies cannot deliver broadly useful agents, then they may be unable to deliver on their promises of an A.I.-powered future.

Since most of us complete computer tasks by pointing and clicking, an A.I. that can “join the workforce” probably needs to know how to use a mouse—a surprisingly difficult goal. The Times recently reported on a string of new startups that have been building “shadow sites”—replicas of popular webpages, like those of United Airlines and Gmail, on which A.I. can analyze how humans use a cursor. In July, OpenAI released ChatGPT Agent, an early version of a bot that can use a web browser to complete tasks, but one review noted that “even simple actions like clicking, selecting elements, and searching can take the agent several seconds—or even minutes.” At one point, the tool got stuck for nearly a quarter of an hour trying to select a price from a real-estate site’s drop-down menu.

https://archive.ph/TcU5B
 
Last edited:
In a sane world the technology would have been flagged as being fundamentally flawed a few years ago and either dust binned entirely, or put on the shelf until renewable power was cheap and plentiful enough that it's wide usage wouldn't collapse society, and the halluncination issue was dealt with.

But we don't live in a sane world. Our tech industry is mostly the theft/misuse of people's data and pumping the valuation of vaporware firms to the fucking moon for shareholder gain , so here we are.
 
In a sane world the technology would have been flagged as being fundamentally flawed a few years ago and either dust binned entirely, or put on the shelf until renewable power was cheap and plentiful enough that it's wide usage wouldn't collapse society, and the halluncination issue was dealt with.

But we don't live in a sane world. Our tech industry is mostly the theft/misuse of people's data and pumping the valuation of vaporware firms to the fucking moon for shareholder gain , so here we are.

part of is just regular human bias...AI was developed by programmers and it's really good at doing those tasks so the assumption was that it could do easier non code related stuff...well it can't
 
part of is just regular human bias...AI was developed by programmers and it's really good at doing those tasks so the assumption was that it could do easier non code related stuff...well it can't

Even within that bias though...it's kind of not that good, but they assumed it would get really really good because it started off kind of okay. This is where all of it "it will replace all of the coders!" stuff came from, where all it's been able to replace professionally is a small percentage of juniors.

The claims that the models are good at coding largely comes from the coding competitions they've created to put it up against human coders that they have to game the shit out of the rules to benefit the AI models in. The models win, but in most/all cases, the rules were designed in a way that benefits a system that can iterate 1000's of times in the time it takes a human to do it the right way once. Oh, the model got it right on it's 500th attempt faster than it took the human to do it right once? Amazing!!! I think most people can pretty quickly imagine how shit a process that would be to design a business around.

The other criticism I've seen from professionals is that the code is often bloated, inelegant, and shit even when it works so isn't suitable for a lot professional applications without having human coders go back over it and trim it significantly. Which is a problem that tracks with issues it has outside of coding. It's not that it can't do stuff, it just can't do it reliably with autonomy, and requires as much human input to correct it as it would have taken to do it without it in the first place.

Where it seems like it's really good at code is to non and novice coders who can use it to do shit that they would be unable to do it without the assistance of an LLM. It's legit amazing for that because bad code that works is infinitely better than no code. I use it to create bad python scripts that help my workflow pretty regularly now. The scripts work after some fucking about and that's all I give a shit about, but I did quiz up a subreddit with the script it gave me when trying to fix something that wasn't working and the consensus was that it was twice as long as it needed to be and made errors that any trained coder wouldn't.
 
I've actually been doing some thinking and i might have figured out some good "big idea" /non-$$$ uses for current AI. But that's just a theory i haven't tested yet.
 
Even within that bias though...it's kind of not that good, but they assumed it would get really really good because it started off kind of okay. This is where all of it "it will replace all of the coders!" stuff came from, where all it's been able to replace professionally is a small percentage of juniors.

The claims that the models are good at coding largely comes from the coding competitions they've created to put it up against human coders that they have to game the shit out of the rules to benefit the AI models in. The models win, but in most/all cases, the rules were designed in a way that benefits a system that can iterate 1000's of times in the time it takes a human to do it the right way once. Oh, the model got it right on it's 500th attempt faster than it took the human to do it right once? Amazing!!! I think most people can pretty quickly imagine how shit a process that would be to design a business around.

The other criticism I've seen from professionals is that the code is often bloated, inelegant, and shit even when it works so isn't suitable for a lot professional applications without having human coders go back over it and trim it significantly. Which is a problem that tracks with issues it has outside of coding. It's not that it can't do stuff, it just can't do it reliably with autonomy, and requires as much human input to correct it as it would have taken to do it without it in the first place.

Where it seems like it's really good at code is to non and novice coders who can use it to do shit that they would be unable to do it without the assistance of an LLM. It's legit amazing for that because bad code that works is infinitely better than no code. I use it to create bad python scripts that help my workflow pretty regularly now. The scripts work after some fucking about and that's all I give a shit about, but I did quiz up a subreddit with the script it gave me when trying to fix something that wasn't working and the consensus was that it was twice as long as it needed to be and made errors that any trained coder wouldn't.
TLDR: Mindz ends every anti-AI monologue with a story about how he uses AI successfully
 
TLDR: Mindz ends every anti-AI monologue with a story about how he uses AI successfully

Mindz has never claimed that it's useless, just that it's use cases aren't what they were expected to be and aren't commercially viable in a way to explain a multi trillion dollar investment.

Yes, Chatgpt saves me from going on fiverr and spending 40 bucks on having someone write a simple python script for me.
 
I've actually been doing some thinking and i might have figured out some good "big idea" /non-$$$ uses for current AI. But that's just a theory i haven't tested yet.


giphy.webp
 
Back
Top