Comment by alfalfasprout | Hacker News Reader

Comment by alfalfasprout | Hacker News Reader

It's a tool that isn't going away. But even Yann Lecun has pointed out that "A house cat has way more common sense and understanding of the world than any LLM".

That's fundamentally the real issue. LLMs are far from even animal level "cognition" in their ability to solve novel problems. A lot of people seem to vastly underestimate the number of novel problems they solve on a regular basis.

In practice, what we're seeing now is something akin to the first wave of "ai hype" that led VCs into a spending frenzy years ago. Some of it did materialize. Most of it didn't.

LLMs and their related bretheren are pretty cool. They're very useful for solving routine problems. There's a lot of "routine" work out there in the world and it will probably replace some of those jobs.

But just like CNNs led to a frenzy of innovation in image recognition and LSTMs pushed NLP-related use cases forward this too will hit a wall.

If I were a betting man I think the main industries LLMs will disrupt are:

1. Search. ChatGPT is decent at summarizing stuff which will make searching more natural. 2. QA/Testing. A lot of this work is pretty repetitive and manual and LLM assistants can generate skeleton code for tests, etc. pretty well. 3. "basic" programming jobs that use "frameworks" to slap together simple apps/websites. A lot of this is repetitive gluing of stackoverflow code already so ChatGPT will make these folks considerably more productive. I'm not sure that these jobs will even disappear either.

I'm surprisingly less bullish on ChatGPT replacing marketing folks. It may be used to generate copy but the risks of putting out brand-damaging copy or a marketing campaign that doesn't make sense are too high.

It's pseudo-understanding, but pseudo-understanding can still be useful, as we see with, say, image classification.

100%

I'm just noting that we're currently in a period where suddenly a lot of folks without the understanding of its limitations are experiencing something AGI-like for the first time. It feels a lot more competent than it could ever be.

Once the limitations of what LLMs are really "competent" at are well-established we'll likely see a huge surge of use cases based on them.

A lot of folks are too quick to claim the death of the knowledge worker though. Folks working in the ML space understand there's a much longer road ahead than we've even traveled down so far.

There's no doubt about that, but it is somewhat limited to cases where "close enough" is actually close enough (or it should be... the worst case is when these sorts of technologies are inappropriately deployed in scenarios where the level of certainty they can give is not enough).

> It may be used to generate copy but the risks of putting out brand-damaging copy or a marketing campaign that doesn't make sense are too high.

My girlfriend works for a certain AI marketing (i.e. copywriting) company that services large customers like Dropbox and Humana. They create and test headlines for email, ads, direct mail, whatever.

She’s an excellent and creative writer, so her job is to help validate and train the models by analyzing the copy it generates and setting up tests. Unfortunately, people like her are creating 99% if the marketing copy at the company and everyone (including the CEO) is in denial about it. The copy just isn’t good or violates brand guidelines.

And because the company is in denial, they’re not investing in more people in her role nor replacing people who leave. As they get overworked, more people leave, and they’re losing accounts because the deniers believe the AI is doing all the work.

It’s pretty ridiculous. I guess they want that SaaS-AI multiplier. But really it’s just a marketing agency.

I honestly find animals are generally very smart. Maybe because they are living things, evolved and born to be survived.

Thank you, I dislike this lazy connection between animals and stupidity. It feels like a verbal tick of some sort thrown around when reaching for an example

But that's exactly my point. Animals are smart. People are underestimating truly how far we have to go to match the intelligence of even the "least intelligent" animals.

> but the risks of putting out brand-damaging copy

There are ways to mitigate that, like slow initial roll-out to a small population. Or something.

>> It's a tool that isn't going away.

Like "Resistance is futile"? That sort of thing?

Reply on news.ycombinator.com