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

Taste in the age of AI and LLMs

rajnandan.com

Post reply on HN
Print view View original

161–170 of 230 posts

Re: Taste in the age of AI and LLMs

#161

Post by subhobroto »

CharlieDigital wrote: post #133
subhobroto wrote: post #122

Earlier quoted context omitted.

> Instead, the right train of thought is: "what would perfect code look like?" and then meticulously describe to the LLM what "perfect" is to shape every line that gets generated. I don't think there's perfect code. Code is automation - it automates human effort and humans themselves have error, hence not perfect. So as long as code meets or exceeds the human output, it's "good enough" and meets expectations. That's…

> I don't think there's perfect code Note I used "perfect" in my text. In this context, meaning it passes human PR reviews following our standard guidelines with minimal feedback/correction required. > So as long as code meets or exceeds the human output, it's "good enough" and meets expectations. That's what a typical customer cares about. Why settle for this when "perfect" is "free"? I understand this dichotomy whe…

> There really is no cost difference between writing shitty code and "perfect" code now; its just a matter of how good you are at describing "perfect" to the LLM.

The consequent is at odds with the antecedent. It's a performative contradiction (if the output were truly "free", the skill of the operator would be a zero-value variable - yet, by requiring skill, you acknowledge a cost) as I prove below

> The cost of "perfect" is only perhaps a few fractions of a cent higher than shitty.

Is your cost model accounting for the cost of specification, of review and additional cycles required if review fails or the specification itself needs to be adjusted?

> If we simply tell the agent these rules ahead of time, we get "perfect", consistent code each time

No, in the simplest case, your cost of perfection is simply moving up the chain of abstraction from implementation (coding) to design and specification. In reality it also splits and moves a part of that cost downstream to verification.

This isn't some special, magical insight I have, I'm reiterating Tesler's Law right back to you.

I also encourage you to read software history - for decades it has been trivial to split out perfectly working CRUD from an ER and UML diagram, no LLM necessary. The insight is understanding why we continue to hire cheap human labor to spit out CRUD instead of using those tools.

The cost of software is, and always has been, in the figuring out the intent, not the generation of syntax.

I wish pg was more active on HN - I expect this is one of the reasons why he wanted founders to have and share the painpoints of their (potential) customers. Figuring out the intent is expensive. Mistake the intent and the best case scenario is a pivot.

Re: Taste in the age of AI and LLMs

#162

Post by freediddy »

Jensson wrote: post #134
hellisothers wrote: post #117

Earlier quoted context omitted.

Aren’t you just making their point stronger? Effort is what is being replaced here, with some taste and a pile of AI (formerly effort) you can go to the moon.

But you still need effort, its not only taste. "Only" means you can do it with no effort.

At some point, just an idea will be enough for your Neurolink to spawn an agent to create 1000 different versions of your idea along with things that mimic your tendencies. There will be no effort, only choice.

Re: Taste in the age of AI and LLMs

#163

Post by edgyquant »

stronglikedan wrote: post #87
danielvaughn wrote: post #34

Disagree with the overall argument. Human effort is still a moat. I've been spending the past couple of months creating a codebase that is almost entirely AI-generated. I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out. There's some truth in there that judgement is as important as ever, t…

> I've gotten way further than I would have otherwise at this pace, but it was still a lot of effort, and I still wasted time going down rabbit holes on features that didn't work out. By the time I'm done learning about the structure of the code that AI wrote, and reviewing it for correctness and completeness, it seems to be as much effort as if I had just written it myself. And I fear that will continue to be the re…

Well that is not how anyone is doing agentic coding though. That sounds like just a worse version of traditional coding. Most people are building test suites to verify correctness and not caring about the code

Re: Taste in the age of AI and LLMs

#164

Post by evdubs »

everybodyknows wrote: post #125
johnfn wrote: post #49

It is profoundly ironic that this article is AI generated.

Trying to bring my nose for AI up to standard -- care to share what you're smelling? For me it's: - Short, declarative sentences, stating grandiose yet vague claims, in a high school vocabulary: "Taste becomes useful when it moves from vibe to diagnosis." - Absence of references (let alone web links) to real-world examples. - Em-dashes, gone. No semicolons, but 23 full colons. As instructed by prompt?

There must be a table with three columns and 4-6 rows.

Re: Taste in the age of AI and LLMs

#165

Post by gwern »

gmaster1440 wrote: post #148
gwern wrote: post #144

Earlier quoted context omitted.

They do improve, but the general creativity and sparkle we see with increasing scale comes mostly from scaling up pretraining/parameter-size, so it's quite slow and expensive compared to the speed (and decreasing cost) people have come to take for granted in math/coding in small cheap models. Hence the reaction to GPT-4.5: exactly as much better taste and discernment as it should have had based on scaling laws, yet r…

I think we're basically agreeing here. Your point (if I'm reading it right) is that taste and discernment do scale, but the gains come through pretraining/parameter scaling, which is slow and expensive compared to the fast, cheap wins in math/coding from smaller models. So taste is more of a lagging indicator of scale. it improves, but it's the last thing people notice because the benchmarkable stuff races ahead. Whi…

My point is more that since you can expect taste's commoditization to lag behind for deep fundamental reasons, then taste does serve as a moat. Just perhaps a weaker one than one would naively expect, and where you will have to frantically keep investing in it to stay ahead of the LLMs slowly catching up, as opposed to a permanent lock-in you can lazily monopolistically coast on indefinitely. (I'm reminded of Neal Stephenson's La Brea tarpit analogy for open source vs proprietary software in _In The Beginning was the Commandline_.)

Re: Taste in the age of AI and LLMs

#166

Post by liuliu »

acuozzo wrote: post #114
liuliu wrote: post #111

Earlier quoted context omitted.

I am not sure you set it up right. Did you have a runnable WolframLanguage file so it can compare results? Did you give it H100 / H200 access to compile and then iterate? My experience is that once you have these two, it does amazing kernel work (Codex-5.4).

> Did you have a runnable WolframLanguage file so it can compare results? Yes. > Did you give it H100 / H200 access to compile and then iterate? Yes via Lambda.ai. Also, FWIW, I run claude with --dangerously-skip-permissions and codex with the equivalent flag. > it does amazing kernel work (Codex-5.4) Specifically with WGMMA + TMA? --- Once TMA gets involved both Claude and Codex spin endlessly until they dump TMA fo…

That is informative, thanks! Yes, I observe the same thing as the model tends to give up (like you said, "dump TMA for a slower fallback") and needs active steering to get good results. But it indeed works further than one-shot from Chat interface and knows much more about profiling / kernel coding than these.

Re: Taste in the age of AI and LLMs

#167

Post by skydhash »

freediddy wrote: post #162
Jensson wrote: post #134

Earlier quoted context omitted.

But you still need effort, its not only taste. "Only" means you can do it with no effort.

At some point, just an idea will be enough for your Neurolink to spawn an agent to create 1000 different versions of your idea along with things that mimic your tendencies. There will be no effort, only choice.

Deciding between 1000 different versions is a lot of effort IMO. With manual coding, you’re mostly deciding one decision point at a time, which is easier when you think about it. It just require foresight which comes from experience

Re: Taste in the age of AI and LLMs

#169

Post by hackable_sand »

_2fnr wrote: post #13

[flagged]

I guess so

If your goal is to turn a profit

If you want to make something that is true then you can't fail. You don't need any defensive structures because the everyone else just spins in lies.

People only need moats when they are insecure, by definition

The most truthful artists don't need that.

Post reply on HN