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

The Generative AI Con

wheresyoured.at

Post reply on HN
Print view View original

131–140 of 503 posts

Re: The Generative AI Con

#131

Post by Lio »

simonw wrote: post #5

> When you put aside the hype and anecdotes, generative AI has languished in the same place, even in my kindest estimations, for several months, though it's really been years. The one "big thing" that they've been able to do is to use "reasoning" to make the Large Language Models "think" [...] This is missing the most interesting changes in generative AI space over the last 18 months: - Multi-modal: LLMs can consume…

> LLMs can consume images,

Not very well in my experience. Last time I checked ChatGPT/DALL-E couldn't understand the its own output to know that what it had drawn was incorrect. Nor could it correct mistakes that were pointed out to it.

For example, I ask it to draw an image of a bike with rim brakes it could not, nor could it "see" that what was wrong with the brakes that it had drawn. For all intents and purposes it was just remixing the images it had been trained on without much understanding.

Re: The Generative AI Con

#132

Post by parodysbird »

NewUser76312 wrote: post #24

Hey I have my gripes with the landscape, certainly, but this is just too much. > It sure is! But it doesn't really prove anything other than that people are using the single-most-talked about product in the world. By comparison, billions of people use Facebook and Google. I don't care about this number! > User numbers alone tell you nothing about the sustainability or profitability of a business, or how those people…

This is not a criticism he really mentions explicitly, but the issue I see with the valuation is that these products cannot be used in a production system by a responsible engineer. As in, I can't have LLMs autonomously plugged in as part of my product. The failure modes are not ever going to be predictable enough. Now, Microsoft will probably be able to charge a fortune at enterprise level, and managers will dream of a day that the LLMs can replace all those weirdo devs at their company, but that'll stay a dream.

All of the valuable uses are personal. It makes me personally feel more productive at work. It helps me personally understand some topic better. It gives me an idea in a personal project.

That's all really cool, but that is not what the valuation is about. The valuation is about a false science fiction and hype bubble about agentic this or that or AGI or whatever, and this is driving very questionable decisions for wasting possibly trillions of dollars and tons of energy.

The plus side is that there is some really cool personally useful tech here, and we will probably end up with very good open source implementations and cheaper used GPUs once the bubble bursts.

Re: The Generative AI Con

#133

Post by cess11 »

spaceman_2020 wrote: post #71
idonotknowwhy wrote: post #25

Earlier quoted context omitted.

Yeah I'm surprised by all the negativity as well. I'm listening to the post right now (using xtts-v2 finetuned on a voice I like lol). Sounds like these companies are overvalued / over hyped. Maybe they are / some of these companies go the way of myspace, but LLMs are incredibly useful for me. I'm able to do a so much more using LLMs (Mistral-Large, Qwen2.5 and R1 locally, Claude via API) than without them. I have to…

Personally, I've found DeepSeek R1 to be a profoundly good model for thinking through problems across fields. I had a complex finance situation that I was struggling with, both from a mathematical/taxation perspective and a personal psychological finance hangup. I spent a few good hours talking to it through everything and had a mental breakthrough. To get the same kind of insight, I would have to pay a financial adv…

Did you run this locally?

Re: The Generative AI Con

#134

Post by parodysbird »

CamperBob2 wrote: post #60
tiborsaas wrote: post #59

> And when this all falls apart — and I believe it will — there will be a very public reckoning for the tech industry. I shocked he really believes it in his closing thought. Maybe his rant would be bit more digestible if it contained sections with: "here's what I tried and it did not work". But that would make it not a rant but actual research with value.

I'll save him the trouble of writing it: "I asked ChatGPT 3.5 to write a large, underspecified chunk of code a couple of years ago, and it didn't work the first time, unlike the code that I write. This whole 'AI' business is an elaborate scam."

He's not a developer. He's really talking about consumer tech.

Re: The Generative AI Con

#136

Post by paulgb »

cmdtab wrote: post #17
paulgb wrote: post #11

> And even then, we still don't have a killer app! There is no product that everybody loves, and there is no iPhone moment! I would strongly argue that coding assistants are AI’s first killer app. Copilot, Cursor, Windsurf etc.

Are they? I find Agentic mode on most editors barely useful. Autocomplete and inline editing is great though. To use these tools properly, you need to know how to build the same thing precisely.

I think autocomplete alone would be enough to make coding a killer app for AI.

I agree the tools are overhyped for allowing non-developers to write code. It’s not (today) a replacement for a dev agency that takes a set of requirements and runs with it, it’s a replacement for a junior developer who you need to micromanage a bit. But that’s still a boon to productivity!

Re: The Generative AI Con

#137

Post by enpera »

lukevp wrote: post #12

I’m a little shocked at how much negativity there is around LLMs among developers. It’s a new tool that requires some learning, and it’s sometimes not so great, but if you’ve used an IDE with real coding assistance built in (eg. VS Code in Edit with Copilot mode - NOT Chat mode, using Claude 3.5), it’s honestly not much worse than a junior dev and 100x faster. And if the code is bad you throw it away and try again 10…

No matter how widespread Copilot becomes, it won't make OpenAI profitable, nor will it enable Sam Altman or Jensen Huang to complete their apps.

Re: The Generative AI Con

#138

Post by heresie-dabord »

jayd16 wrote: post #18
lukevp wrote: post #12

I’m a little shocked at how much negativity there is around LLMs among developers. It’s a new tool that requires some learning, and it’s sometimes not so great, but if you’ve used an IDE with real coding assistance built in (eg. VS Code in Edit with Copilot mode - NOT Chat mode, using Claude 3.5), it’s honestly not much worse than a junior dev and 100x faster. And if the code is bad you throw it away and try again 10…

Unreliable tools are utterly exhausting. > not much worse than a junior dev and 100x faster. Is there a greater hell than this!?

> a junior dev and 100x faster. > Is there a greater hell than this!?

Yes — junior management using LLMs and 100x more cocksure.

Re: The Generative AI Con

#139

Post by petesergeant »

Article is a motte-and-bailey[0] argument.

Bailey (the clickbait): “Generative AI is a con!!”

Motte (narrow defendable argument): OpenAI and Anthropic have not shown that building a proprietary model and selling inference is a sustainable business.

0: https://en.m.wikipedia.org/wiki/Motte-and-bailey_fallacy

Re: The Generative AI Con

#140

Post by jcgrillo »

danroblew wrote: post #84
simonw wrote: post #5

> When you put aside the hype and anecdotes, generative AI has languished in the same place, even in my kindest estimations, for several months, though it's really been years. The one "big thing" that they've been able to do is to use "reasoning" to make the Large Language Models "think" [...] This is missing the most interesting changes in generative AI space over the last 18 months: - Multi-modal: LLMs can consume…

Cost as in, cost to you? Or cost to serve? If the cost-to-serve is subsidized by VC money, they aren't getting cheaper, they're just leading you on.

> Cost as in, cost to you? Or cost to serve?

This. IIUC to serve an LLM is to perform an O(n^2) computation on the model weights for every single character of user input. These models are 40+GB so that means I need to provision about 40GB RAM per concurrent user and perform hundreds of TB worth of computations per query.

How much would I have to charge for this? Are there any products where the users would actually get enough value out of it to pay what it costs?

Compare to the cost of a user session in a normal database backed web app. Even if that session fans out thousands of backend RPCs across a hundred services, each of those calls executes in milliseconds and requires only a fraction of the LLM's RAM. So I can support thousands of concurrent users per node instead of one.

Post reply on HN