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

Let's talk about LLMs

b-list.org

Post reply on HN
Print view View original

101–110 of 201 posts

Re: Let's talk about LLMs

#101

Post by tptacek »

threecheese wrote: post #88
kasey_junk wrote: post #80

Earlier quoted context omitted.

If soc audits are driving your development process you are doing it backwards. And _certainly_ a time is coming when just using the llm will be soc compliant.

I’d think any company big enough or working in certain markets which has a Compliance Officer cares about this; regulations are a legitimate business risk, and software integration contracts have security control compliance requirements which very much impact the sdlc. Would you have the same reaction to requiring an approval for a production deployment? That’s driving the development process. —- Also jfc I need to c…

SOC2 is generally regarded as a joke and has in fact almost nothing to do with software resilience even on its own terms.

Re: Let's talk about LLMs

#102

Post by gerdesj »

aspenmartin wrote: post #99
gerdesj wrote: post #91

Earlier quoted context omitted.

"paradigm shift" A paradigm shift is an earth shattering, very important change - a complete change in thinking etc. LLMs are not that. They are simply some pretty new tools. Nice tools but they will whip off your metaphorical thumb just as quickly as a miss-used table saw. You'll note that you mention "engineers are offloading": that's not a paradigm shift. That's a bunch of engineers discovering a better slide rule…

I would argue LLMs are possibly the largest paradigm shift the world has ever seen, and we are only at the beginning. The entire scaffolding and structure of programming is in the process of changing — coding has moved to orchestration and testing and governance of how to manage and productionalize code that has surpassed the capacity of human review. If this sounds melodramatic it’s likely that it hasn’t fully taken…

[How did you bang out this: — on your keyboard? Why did you decide to use backticks and 66/99 for quotes - nice but its not you is it?]

Engage as a person, please.

Re: Let's talk about LLMs

#103

Post by mfro »

imiric wrote: post #43
mfro wrote: post #4

I think you're misunderstanding the paradigm shift completely -- AI does not just generate code N(x) more quickly. It thinks N(x) faster, it researches N(x) faster, it tests N(x) faster. There are hundreds of tasks that you'll find engineers are offloading to AI every day. The major hurdle right now is actually pivoting LLMs from just generating code: integrating those tasks into workflows. This is why tool-use and a…

> The major hurdle right now is actually pivoting LLMs from just generating code: integrating those tasks into workflows. Funny, I thought that the major hurdle is improving accuracy and reliability, as it's always been. Engineering is necessary and useful, but it's a much simpler problem, which is why everyone is jumping on it.

As much as that’s true it’s clear a huge amount of people have accepted the current state and are working around it, successfully(in terms of ticking an executive’s checkbox) in a lot of cases. And it’s worth considering we’re seeing strong strides outside of model quality in the tooling and integration

Re: Let's talk about LLMs

#105

Post by furyofantares »

Considering some project I'm able to accomplish today by hand, it is much larger than it was 25 years ago, but back then it would have mostly been inherent difficulty and now the work is mostly accidental difficulty. The libraries available have solved almost all of the inherent difficulty for many small (solo-dev) projects, and so for many small projects I already know every I need to know to do it (even though I don't know everything I would have needed to do it 25 years ago.) All that's left is doing it.

Re: Let's talk about LLMs

#106

Post by simonw »

leptons wrote: post #84
kelnos wrote: post #27

>> Within just this group the ratios between best and worst performances averaged about 10:1 on productivity measurements and an amazing 5:1 on program speed and space measurements! > (although I’m personally skeptical of the “10x programmer” concept, the software industry overall does seem to accept it as true) To be fair, this statement from Brooks doesn't entirely match with the "10x programmer" we talk about. My…

There's no such thing as a "10x" programmer, and anyone who uses it doesn't know what they're talking about. 10x relative to what exactly? It's not a statement grounded in any kind of reality.

10x makes sense only in terms of specific technology platforms.

I'm a 10x programmer at building Django apps compared to a developer who has never worked with Django before.

Someone who developers against WordPress on a daily basis will easily 10x my own attempts at building things on that platform.

Re: Let's talk about LLMs

#107

Post by aspenmartin »

strange_quark wrote: post #69
aspenmartin wrote: post #41

Earlier quoted context omitted.

You say this as though performance has not followed a very clear and extremely rapid improvement in a startlingly short amount of time. You’re definitely right that people adopt agentic workflows and are disappointed or worse, but the point is the disappointment has already reduced substantially and will continue to do so. We know this because we know the scaling laws, and also because learning theory has been around…

What rapid improvement has occurred, because in this six month AI coding fever dream we've been living in, I really haven't seen anything new in awhile, both in terms of new ideas for AI coding or in new consumer products or services. I'll give you the coding harnesses themselves are better because that was a new product category with a lot of low-hanging fruit, but have the models actually improved in a way that isn…

I’m going to parrot back what you’re saying and you tell me if I’m getting close

- AI coding is a disappointing fad (“fever dream?”). - that has not made meaningful progress in…6 months? - coding harness is improving - model improvements are lies: it’s just businesses “benchmaxxing” and misleading people. Real performance has not meaningfully improved - “opus 4.7 is a dud” - 5.5 suffering from “system collapse” (I’ve never heard this term before)

Since you asked and I assume you are rational and really are interested to know:

- we have many measures of performance and have studied how one particularly important but unintuitive measure (pertaining perplexity) scales with data, compute, and model size. These laws continue to hold and have satisfying theoretical origins.

- whatever the scale of 5.5, consider we have far more room to go on the scaling front. Probably another 2-3 orders of magnitude before we hit limiting bottlenecks.

- that’s also fine because scaling is only part of the puzzle. RL on verifiable rewards is virtually guaranteed to get you optimal performance and that’s the entirety of the excitement around coding agents

- while you are right about benchmarks and measurement science having a ton of weaknesses, they are not at all garbage. There are probably around 40,000 benchmarks in the literature (this is not a made up number by the way it really is around that many). Epoch made a great composite measure using good stats (IRT) called their epoch capability index, METR has done and redone their time horizon measure and it holds up beautifully. There is a ton of signal in many benchmarks and they all tell a pretty compelling story.

- additionally, this is not some unknowable thing. It strikes me as odd that people’s prior on HN a lot of time is “it’s all dumb rich people putting way too much dumb money in this”. Sorry but the world is not that dumb. Trillions of CapEx is usually pretty rationally allocated. And it is!

- why? Because this is already known what happens when you do what we’re doing. When you have a verifiable reward system, have a certain amount of compute available, have seed data to get you to where you can do RL, you will be almost guaranteed to get superhuman performance

Re: Let's talk about LLMs

#108

Post by aspenmartin »

gerdesj wrote: post #102
aspenmartin wrote: post #99

Earlier quoted context omitted.

I would argue LLMs are possibly the largest paradigm shift the world has ever seen, and we are only at the beginning. The entire scaffolding and structure of programming is in the process of changing — coding has moved to orchestration and testing and governance of how to manage and productionalize code that has surpassed the capacity of human review. If this sounds melodramatic it’s likely that it hasn’t fully taken…

[How did you bang out this: — on your keyboard? Why did you decide to use backticks and 66/99 for quotes - nice but its not you is it?] Engage as a person, please.

I typed this out, character by painful human character, on an iPhone. It is indeed me!

Re: Let's talk about LLMs

#109

Post by zouhair »

marcus_holmes wrote: post #94

I was waiting for the "so I tried coding something with an LLM myself, and I found..." paragraph. But apparently the author never did try it, or at least if they did, they didn't write about it. This is a very academic approach to the subject - read what other people have written about it without ever doing it yourself. Study what someone said about LLM coding 50 years ago, before they were even invented, to see what…

The problem I have with it is the price (I am not talking about the money). I don't know if the price is worth it. For example we are literally witnessing the death of the personal computers, it will soon become a rich people's hobby. I don't know how the whole Free Software/Open source will survive that.

At best we will end up not owning nothing, not even the programming skills as everyone will be at the mercy of AI companies for their coding.

We are still in the honey moon phase of AI coding, I have a very pessimistic view of the future.

Re: Let's talk about LLMs

#110

Post by aspenmartin »

cyclopeanutopia wrote: post #56
aspenmartin wrote: post #41

Earlier quoted context omitted.

You say this as though performance has not followed a very clear and extremely rapid improvement in a startlingly short amount of time. You’re definitely right that people adopt agentic workflows and are disappointed or worse, but the point is the disappointment has already reduced substantially and will continue to do so. We know this because we know the scaling laws, and also because learning theory has been around…

Perhaps you are confusing performance with instability?

No. Time horizon I’m talking about spans years. “We don’t know” is just wrong, we’ve had scaling laws for many years and they continue to hold up. Benchmarks, in all their ugliness, tell a consistent story.
Post reply on HN