Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LLMs aren't remotely like compilers or power tools

blainehansen.me

Post reply on HN
Print view View original

41–50 of 55 posts

Re: LLMs aren't remotely like compilers or power tools

#41

Post by jurgenburgen »

throw310822 wrote: post #36
bluefirebrand wrote: post #31

Earlier quoted context omitted.

It's a clear illustration that these systems are not oracles. No matter how much they've improved since then, we should assume they are still capable of making hilariously big mistakes. Only now the obvious mistakes are fixed and the big mistakes will be a lot more subtle and a lot harder to catch.

> It's a clear illustration that these systems are not oracles. Of course they are not oracles. Oracles belong to magic, these things are real. I really can't understand how people seem to expect "intelligence" to be deterministic and infallible and uniform across domains when all examples we have of it (that is, us) are anything but.

> I really can't understand how people seem to expect "intelligence" to be deterministic and infallible and uniform across domains when all examples we have of it (that is, us) are anything but.

I’m sure there are people who think putting glue on a pizza is acceptable but we’re not hiring them to bake pizzas for us.

When I search and get an AI summary it’s either correct or useless. I don’t care how fallible humans are, I’m not asking them to summarize my Google results.

Re: LLMs aren't remotely like compilers or power tools

#42

Post by jdlshore »

Jemaclus wrote: post #39
inigyou wrote: post #22

Earlier quoted context omitted.

Evidence from listening to slopcoders speak says they feel it's night and day. Evidence from observing reality shows a 20% speed increase at best and 50% speed decrease at worst and almost always a significant reduction in quality.

I was also skeptical, but I started measuring quality (I'm using rework as a rough metric) and quantity (PR throughput), and I've found that, on average, LLMs are 60% worse at writing code (meaning they rework at a rate 60% higher than humans) but my team produce PRs about 2.4x faster. So objectively my team's numbers don't match up with yours: our speed is about 2.5x increase _on average_ and a 60% reduction _on ave…

Quality is more about maintainability, IMO, which comes down to consistency and clarity of design, sensible abstractions, cohesion, decoupling, etc. I don’t think a rework metric tells you that.

Re: LLMs aren't remotely like compilers or power tools

#43

Post by david-gpu »

yifanl wrote: post #40
david-gpu wrote: post #37

Earlier quoted context omitted.

I am just saying that people have been making these sorts of shallow dismissive comments about people who use different tooling since forever. I remember a college professor snided at students who didn't know how to use vi. In the same way that most people are no longer proficient in assembly, but it has not spelled doom for software engineering, some folks are now using AI tools, and it will probably be fine. The da…

Its true that things change, but it's also true that there will eventually be a day when the last programmer is hired.

We don't have very many coopers left, either. Sometimes technological progress makes some occupations obsolete.

I don't think software engineering will disappear, but it could change a lot, potentially. More understanding the needs of the end consumer, more writing specifications, less raw coding and debugging. Maybe fewer programmers will be needed, and that would not be the end of the world.

Re: LLMs aren't remotely like compilers or power tools

#44

Post by thewhitetulip »

vrighter wrote: post #13

i recently heard a colleague telling another "vscode can't really help you when writing yaml files". I said that the yaml lsp supports loading a json schema, all you need is just one comment at the top to point the lsp to it, then you get reliable error reporting with red squigglies on every single keystroke. He just shrugged and said "oh i just use ai". What takes me one keystroke takes him a full minute. And he did…

I've seen people comitting to git using AI lol

Re: LLMs aren't remotely like compilers or power tools

#45

Post by bluefirebrand »

throw310822 wrote: post #36
bluefirebrand wrote: post #31

Earlier quoted context omitted.

It's a clear illustration that these systems are not oracles. No matter how much they've improved since then, we should assume they are still capable of making hilariously big mistakes. Only now the obvious mistakes are fixed and the big mistakes will be a lot more subtle and a lot harder to catch.

> It's a clear illustration that these systems are not oracles. Of course they are not oracles. Oracles belong to magic, these things are real. I really can't understand how people seem to expect "intelligence" to be deterministic and infallible and uniform across domains when all examples we have of it (that is, us) are anything but.

> I really can't understand how people seem to expect "intelligence" to be deterministic and infallible

We don't really expect this of intelligence

We do sort of expect this of machines

Most people in society use machines and computers expecting very deterministic behavior, within some constraints. A saw cuts wood as long as your blade isn't dull. A computer shows your emails as long as the power and Internet is not out.

Comparably Unconstrained computer AI systems are still extremely new and there is no social norms or expectations around it yet

Re: LLMs aren't remotely like compilers or power tools

#46

Post by throw310822 »

jurgenburgen wrote: post #41
throw310822 wrote: post #36

Earlier quoted context omitted.

> It's a clear illustration that these systems are not oracles. Of course they are not oracles. Oracles belong to magic, these things are real. I really can't understand how people seem to expect "intelligence" to be deterministic and infallible and uniform across domains when all examples we have of it (that is, us) are anything but.

> I really can't understand how people seem to expect "intelligence" to be deterministic and infallible and uniform across domains when all examples we have of it (that is, us) are anything but. I’m sure there are people who think putting glue on a pizza is acceptable but we’re not hiring them to bake pizzas for us. When I search and get an AI summary it’s either correct or useless. I don’t care how fallible humans a…

So you never go to a doctor because (being human and therefore fallible) they might give you wrong information?

Re: LLMs aren't remotely like compilers or power tools

#47

Post by inigyou »

jdlshore wrote: post #42
Jemaclus wrote: post #39

Earlier quoted context omitted.

I was also skeptical, but I started measuring quality (I'm using rework as a rough metric) and quantity (PR throughput), and I've found that, on average, LLMs are 60% worse at writing code (meaning they rework at a rate 60% higher than humans) but my team produce PRs about 2.4x faster. So objectively my team's numbers don't match up with yours: our speed is about 2.5x increase _on average_ and a 60% reduction _on ave…

Quality is more about maintainability, IMO, which comes down to consistency and clarity of design, sensible abstractions, cohesion, decoupling, etc. I don’t think a rework metric tells you that.

That actually is incorrect. Quality is about how well something works.

Re: LLMs aren't remotely like compilers or power tools

#48

Post by jdlshore »

inigyou wrote: post #47
jdlshore wrote: post #42

Earlier quoted context omitted.

Quality is more about maintainability, IMO, which comes down to consistency and clarity of design, sensible abstractions, cohesion, decoupling, etc. I don’t think a rework metric tells you that.

That actually is incorrect. Quality is about how well something works.

There’s two types of quality: external quality, which is how well something works, and internal quality, which is how well something is built.

Re: LLMs aren't remotely like compilers or power tools

#49

Post by blainehansen »

suprjami wrote: post #12

You had me up until you equated LLM usage to unreliable human delegation. No human being is going to genuinely suggest to glue cheese onto a pizza to stop it sliding off. These things aren't people. Stop anthropomorphizing a computer program. They'll make all the mistakes humans can make, as well as a new class of mistakes because they're just fancy next word prediction with no lived experience.

Hey, original author here.

> But fundamentally any metaphor will be inadequate, LLMs are too weird. It doesn't make sense to compare a transformer-based deep neural network that autoregressively generates language tokens based on their embedding in a high-dimensional semantic space to really anything that's ever existed.

You might not have read the full post ha

Re: LLMs aren't remotely like compilers or power tools

#50

Post by blainehansen »

camkego wrote: post #5

I've got to give the author credit, it seems like this is written without an LLM. But, there seems to be a lot that is arguable. For example, the article says "Natural language is the interface of last resort." But, is that actually true? Isn't the AI boom happening because N/L is truly available as an interface? Anyway, please keep writing and asking tough questions.

Hey original author here. Natural language is always easy and available as an interface, but it's extremely low fidelity. It's not easy to come up with better interfaces for tasks, but when we do they tend to be much better (as in more clear, faster, higher quality, etc).
Post reply on HN