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

The Rise of Whatever

eev.ee

Post reply on HN
Print view View original

21–30 of 537 posts

Re: The Rise of Whatever

#21

Post by beckthompson »

tptacek wrote: post #18
beckthompson wrote: post #4

Earlier quoted context omitted.

AIs still frequently make up stuff up - there isn't really a way to get out of that. Have they improved a lot in the last six months? 100%! But they still make mistakes its quite common

LLM calls make stuff up. Your compiler can't make things up. An agent iterates LLM calls. When your LLM call makes an API up, your compiler will generate errors. The errors get fed back into the iterative loop. In pretty much ever real case, the LLM corrects, but either way: the result is clear. The code may be wrong, but it shouldn't hallucinate entire APIs.

But just compiling doesn't mean that much and doesn't really solve the core issue of AIs making stuff up. I could hook up a random word generator into a compiler and it also would also pass that test!

For example, just yesterday I asked an AI a question about how to approach a specific problem. It gave an answer that "worked" (it compiled!) but in reality it didn't really make any sense and would add a very nasty bug. What it wrote (It used a FrameUpdate instead of a normal Update) just didn't make sense on a basic level of how the framework worked.

Re: The Rise of Whatever

#22

Post by csomar »

tptacek wrote: post #2

LLM output is crap. It’s just crap. It sucks, and is bad. Still don't get it. LLM outputs are nondeterministic. LLMs invent APIs that don't exist. That's why you filter those outputs through agent constructions, which actually compile code. The nondeterminism of LLMs don't make your compiler nondeterministic. All sorts of ways to knock LLM-generated code. Most I disagree with, all colorable. But this article is based…

> LLM outputs are nondeterministic.

LLM outputs are deterministic. There is no intrinsic source of randomness. Users can add randomness (temperature) to the output and modify it.

> But this article is based on a model of LLM code generation from 6 months ago

There hasn't been much change in models from 6 months ago. What happened is that we have better tooling to sift through the randomly generated outputs.

I don't disagree with your message. You are being downvoted because a lot of software developers are butt-hurt by it. It is going to force a change in the labor market for developers. In the same way the author is butt-hurt that they didn't buy Bitcoin in the very early days (as they were aware of it) and missed the boat on that.

Re: The Rise of Whatever

#24

Post by N_Lens »

tyre wrote: post #5

The author is mad at Stripe and PayPal for banning transactions involving unicorn wieners but this is imposed on them by the backing banks. The reason behind banning adult materials has to do with Puritanism and with the high rates of refunds on adult websites.

There's a reason "Paypal mafia" is in the lexicon.

Re: The Rise of Whatever

#25

Post by tptacek »

csomar wrote: post #22
tptacek wrote: post #2

LLM output is crap. It’s just crap. It sucks, and is bad. Still don't get it. LLM outputs are nondeterministic. LLMs invent APIs that don't exist. That's why you filter those outputs through agent constructions, which actually compile code. The nondeterminism of LLMs don't make your compiler nondeterministic. All sorts of ways to knock LLM-generated code. Most I disagree with, all colorable. But this article is based…

> LLM outputs are nondeterministic. LLM outputs are deterministic. There is no intrinsic source of randomness. Users can add randomness (temperature) to the output and modify it. > But this article is based on a model of LLM code generation from 6 months ago There hasn't been much change in models from 6 months ago. What happened is that we have better tooling to sift through the randomly generated outputs. I don't d…

There hasn't been much change in models from 6 months ago.

I made the same claim in a widely-circulated piece a month or so back, and have come to believe it was wildly false, the dumbest thing I said in that piece.

Re: The Rise of Whatever

#26

Post by tptacek »

beckthompson wrote: post #21
tptacek wrote: post #18

Earlier quoted context omitted.

LLM calls make stuff up. Your compiler can't make things up. An agent iterates LLM calls. When your LLM call makes an API up, your compiler will generate errors. The errors get fed back into the iterative loop. In pretty much ever real case, the LLM corrects, but either way: the result is clear. The code may be wrong, but it shouldn't hallucinate entire APIs.

But just compiling doesn't mean that much and doesn't really solve the core issue of AIs making stuff up. I could hook up a random word generator into a compiler and it also would also pass that test! For example, just yesterday I asked an AI a question about how to approach a specific problem. It gave an answer that "worked" (it compiled!) but in reality it didn't really make any sense and would add a very nasty bug…

I'm not interested in this Calvinball argument. The post we're commenting on makes a clear claim: an LLM hallucinating entire APIs. Not surreptitiously sneaking subtly shitty stuff past a compiler.

This is my problem: not that people are cynical about LLM-assisted coding, but that they themselves are hallucinating arguments about it, expecting their readers to nod along. Not happening here.

Re: The Rise of Whatever

#28

Post by csomar »

beckthompson wrote: post #4
tptacek wrote: post #2

LLM output is crap. It’s just crap. It sucks, and is bad. Still don't get it. LLM outputs are nondeterministic. LLMs invent APIs that don't exist. That's why you filter those outputs through agent constructions, which actually compile code. The nondeterminism of LLMs don't make your compiler nondeterministic. All sorts of ways to knock LLM-generated code. Most I disagree with, all colorable. But this article is based…

AIs still frequently make up stuff up - there isn't really a way to get out of that. Have they improved a lot in the last six months? 100%! But they still make mistakes its quite common

You can improve on that

1. A type-strict compiler.

2. https://github.com/isaacphi/mcp-language-server

LLMs will always make stuff up because they are lossy. In the same way that if I ask you to list the methods for some random object lib you'd not be able to do that; you use the documentation to pull that up or your code-complete companion. LLMs are just getting the tools for that.

Re: The Rise of Whatever

#29

Post by bruce511 »

maegul wrote: post #14
gyomu wrote: post #10

Broadly agreed with all the points outlined in there. But for me the biggest issue with all this — that I don't see covered in here, or maybe just a little bit in passing — is what all of this is doing to beginners, and the learning pipeline. > There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge th…

Agreed! The only silver lining I can see is that a new perspective may be forced on how well or badly we’ve facilitated learning, usability, generally navigating pain points and maybe even all the dusty presumptions around the education / vocational / professional-development pipeline. Before, demand for employment/salary pushed people through. Now, if actual and reliable understanding, expertise and quality is desir…

To be fair, LLMs are just the most recent step in a long road of doing the same thing.

At any point of progress in history you can look backwards and forwards and the world is different.

Before tractors a man with an ox could plough x field in y time. After tractors he can plough much larger areas. The nature of farming changes. (Fewer people needed to farm more land. )

The car arrives, horses leave. Computers arrive, the typing pool goes away. Typing was a skill, now everyone does it and spell checkers hide imperfections.

So yeah LLMs make "drawing easier". Which means just that. Is that good or bad? Well I can't draw the old fashioned way so for me, good.

Cooking used to be hard. Today cooking is easy, and very accessible. More importantly good food (cooked at home or elsewhere) is accessible to a much higher % of the population. Preparing the evening meal no longer starts with "pluck 2 chickens" and grinding a kilo of dried corn.

So yeah, LLMs are here. And yes things will change. Some old jobs will become obsolete. Some new ones will appear. This is normal, it's been happening forever.

Re: The Rise of Whatever

#30

Post by loire280 »

tptacek wrote: post #18
beckthompson wrote: post #4

Earlier quoted context omitted.

AIs still frequently make up stuff up - there isn't really a way to get out of that. Have they improved a lot in the last six months? 100%! But they still make mistakes its quite common

LLM calls make stuff up. Your compiler can't make things up. An agent iterates LLM calls. When your LLM call makes an API up, your compiler will generate errors. The errors get fed back into the iterative loop. In pretty much ever real case, the LLM corrects, but either way: the result is clear. The code may be wrong, but it shouldn't hallucinate entire APIs.

A great solution to this problem, but it doesn't seem like this approach will generalize to problems in other fields, or even to more suble coding confabulations that can't be detected by the compiler or static analysis.
Post reply on HN