July 3, 2026
Goalposts on wheels
It Still Can't Do My Job: Four Years of Moving Goalposts (2022–2026)
AI keeps getting better, but the internet keeps yelling “that still doesn’t count”
TLDR: The article argues that artificial intelligence has clearly improved since 2022, while critics keep changing the test for what would count as impressive. In the comments, people split between “the doubters are in denial” and “the hype is still ahead of reality, and it still makes stuff up.”
The real show here isn’t just the timeline of artificial intelligence getting less embarrassing — it’s the comment-section Olympics that followed every single improvement. In late 2022, people laughed when ChatGPT mangled simple tasks and confidently made stuff up, with Stack Overflow essentially saying, “please stop flooding us with wrong answers.” The mood was smug: call me when it can do real work. Then it started passing exams, writing working game code, and helping real programmers at big companies — and the crowd instantly moved the finish line. Suddenly it was just a toy, just copying, just autocomplete, just boilerplate. Ouch.
That’s where the drama really explodes. One camp says this is the oldest trick on the internet: every time the tools improve, skeptics invent a brand-new standard. One commenter joked that in five years people will still say it’s not true machine intelligence because a human can “ride a unicycle blindfolded better than a robot.” The other camp is not buying the hype at all. Their main complaint? It still makes things up, and the hype machine has been even more ridiculous than the technology itself. Several commenters basically rolled their eyes at billionaire boss predictions of office jobs vanishing “within 6 months,” while others said the bigger issue is simple: if this stuff is so revolutionary, why do the business numbers still feel shaky? Even the little details got roasted, with one reader sniping at the article for using AI-written blurbs at all. In other words: the bots improved, but the internet’s favorite hobby remains arguing about whether that improvement counts.
Key Points
- •The article says ChatGPT’s early coding performance in late 2022 was unreliable enough that Stack Overflow banned AI-generated answers from the site.
- •The article describes GPT-4 as a clear capability improvement over early ChatGPT, including generating a working Snake game and posting stronger exam results.
- •The article highlights the launch of Cognition’s Devin and notes that developer Carl Brown later argued the public demo was curated and misleading in key respects.
- •The article cites Nvidia CEO Jensen Huang’s statement in Dubai that human language is becoming the programming language, while also noting growing developer adoption of GitHub Copilot.
- •The article points to Google’s claim that over a quarter of new code is AI-generated and to Pieter Levels’ AI-assisted 3D flight simulator as examples of broader real-world use.