Hacker News Reader

Comment by lionkor | Hacker News Reader

Still haven't seen any evidence of LLMs displaying any kind of intelligence at all, never have. This goalpost has not moved. They are not building AGI, Google is not building a universal search engine, Meta is not building a metaverse you can live in, Crypto isn't replacing fiat currencies, ...

View on HN

Forget about terminology, if you've been following developments for a while, you can't help but notice consistent improvements, and the new models are very powerful. I don't care if anybody calls it AGI, whatever they are, they will be massively transformative.

They're certainly useful, and the goalposts for what is "intelligent" keep shifting, but I think we're seeing only progress towards more polished bigger LLMs. I don't think that progress implies there's a path from LLMs to something substantially better.

LLMs are still prone to hallucinations, and we're only adding more data and workarounds to make it happen less often. Prompt injections limit their usefulness on untrusted inputs. They can't do precise logic and reasoning, and are too likely to follow memorized patterns instead.

We've got something amazing, way better than what we've had before, but the current architecture is still based on a fuzzy translator. It's hard to say whether this is it, and it's going to plateau at this level for a while, or whether there are more breakthroughs around the corner.

OK. With the best of intentions and real curiosity: How?

How they are better? The o1 models succeed where all other models fail.

I don't know how to go more in depth now, but I use Claude, 4o, and o1, both mini and preview in parallel and o1 both mini and preview succeed in many tasks where the others fail.

Fuller question: You say they are massively transformative. In why way? How is life or business going to be changed by this tool?

Like making data entry easier, extract to JSON. Partner for foreign language practice. Brainstorming. Coding and fixing computer problems. Explaining legalese. Reading electronic chips by the label (ICs, transistors) and telling what they do. Fixing my poor English. Turning a list of ideas into a full fledged article. Explaining papers. Writing personalized stories about the adventures of your pet.

Thanks... none of those sound like they would pay the massive bills of running chatgpt (or anything similar).

We're only two years in. To be frank I'm a loss on how to respond to the skepticism. If something so useful can't make a profit in the future you envision, I don't know what can.

They don't seem that useful to me; perhaps that is why our perspectives differ. They seem, at most, like slight conveniences. But the other half is the cost of the service: regardless of whether something is useful (wide spectrum, there), it has to cover the costs of providing it. It costs a lot to provide a service like chatgpt. So far, I don't see people rushing to pay for it. I wouldn't pay for it, in part because sorting out all the misinformation eliminates a lot of the benefit, from my perspective.

Most things that succeed have a very clear value proposition that justifies their cost.

But I watch, and if your envisioned future arrives, that will be interesting.

I don't remember who said it, but it wasn't some AI hype-rer, "LLMs eliminate 40% of the drudgery in code" - which has certainly been my experience, so, on that specific point, this is far from a slight convenience.

Then you shouldn't pay for internet access either because the web is brimming with misinformation and manipulating content. It's not a GPT-specific problem.

I said they will be, from looking at the trajectory. Programmers are just early adopters.

Reply on news.ycombinator.com