The AI Mirage or Why I Think the Hype Can't Sustain Itself | Hacker News Reader

The AI Mirage or Why I Think the Hype Can't Sustain Itself

louwrentius.com

6 pointslouwrentius11 comments

> Imagine that a systems can't be trusted. That 1+1 isn't always 2. Only 99% of the time.

Easy to imagine. Every human being is such a system- correct very often (but never 100% of the times) in certain domains, tragically wrong in others. Like this guy thinking that "AI hype can't sustain itself", with an argument that makes no sense at all. LLMs don't replace computers and algorithms, they replace human beings. They just need to be more correct than humans in several domains, not infallible. Or even slightly less correct, but substantially cheaper.

LLM by their nature can't really replace people. Companies found this out from experience, as they fired and rehired people multiple times, you've seen the articles.

> LLM by their nature can't really replace people

What LLMs could or couldn't do in 2026 doesn't necessarily say anything about their nature, given that they're a rapidly evolving technology that has made giant progress in the space of a few years.

About the - humans aren't accurate either - 'argument'. Why replace people with something objectively worse and more unpredictable?

They aren't necessarily worse and more unpredictable. IIRC. hallucination levels for the frontier levels are below equivalent average normal human errors, and they might even be more systematic/predictable to some degree.

But even if they were slightly worse and more unpredictable, they are many times faster and cheaper, for some tasks.

That word 'some' does quite a bit of heavy lifting.

Because there is a ton of work that people tend to perform that falls out the scope of that 'some'. And a lot of work requires alignment with other people, and so on. LLMs don't have agency, despite people hyping 'agents'.

> That word 'some' does quite a bit of heavy lifting.

No, I was just indicating I was not a sith.

This post misses a very important point: humans aren't 100% correct either. This means that the bar for being useful (at tasks that humans usually do) isn't perfection, it's human-level correctness. If we can create AI models that make fewer mistakes than humans (which is almost certainly possible, even if not easily or soon), then we will actually need to spend less time double-checking for correctness than we do now.

> We see it with self-driving cars. It's really impressive what is possible. But they aren't true self-driving. A person needs to be sitting at the wheel, keeping their attention focussed on traffic as if they would be driving themselves, so they can intervene when the AI makes an inevitable mistake.

There are self-driving cars all over San Fransisco transporting people on public roads with no human at the wheel. This proves my point: those cars are not perfect, but they are human-level (or close), and that's all that's needed.

Not even "fewer mistakes than humans," just make the same mistakes but a lot faster and that's also good enough reason to keep it around (esp when you add parallelization) just from an efficiency/productivity point of view. There may still be other anti AI arguments, ofc.

All that output needs to be checked, thus why making mistakes faster doesn't seem valuable. How do we determine mistakes are made in the first place. Is it determined by people?

> Is it determined by people?

Could be. Machines don't need to provide all the functions of human beings to replace many workers. You can concentrate the supervision and direction function in a single person and assign the rest of the grunt job to machines. You still have replaced most of your human workers.

Reply on news.ycombinator.com