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

AI is a floor raiser, not a ceiling raiser

elroy.bot

Post reply on HN
Print view View original

101–110 of 218 posts

Re: AI is a floor raiser, not a ceiling raiser

#101

Post by gruez »

jjk166 wrote: post #93
gruez wrote: post #68

Earlier quoted context omitted.

Read my comment again. keyword here is "recent". The second link also expands on why it's relevant. It's best to read the whole article, but here's a paragraph that captures the argument: >The shift in recent economic research supports his observation. Although early studies suggested that lower performers could benefit simply by copying AI outputs, newer studies look at more complex tasks, such as scientific researc…

All of the studies were done 2023-2024 and are not listed in order that they were conducted. The studies showing reduced equality all apply to uncommon tasks like material discovery and debate points, whereas the ones showing increased equality are broader and more commonly applicable, like writing, customer interaction, and coding.

>All of the studies were done 2023-2024 and are not listed in order that they were conducted

Right, the reason why I pointed out "recent" is that it's new evidence that people might not be aware of, given that there were also earlier studies showing AI had the opposite effect on inequality. The "recent" studies also had varied methodology compared to the earlier studies.

>The studies showing reduced equality all apply to uncommon tasks like material discovery and debate points

"Debating points" is uncommon? Maybe not everyone was in the high school debate club, but "debating points" is something that anyone in a leadership position does on a daily basis. You're also conveniently omitting "investment decisions" and "profits and revenue", which basically everyone is trying to optimize. You might be tempted to think "Coding efficiency" represents a high complexity task, but the abstract says the test involved "Recruited software developers were asked to implement an HTTP server in JavaScript as quickly as possible". The same is true of the task used in the "legal analysis" study, which involved drafting contracts or complaints. This seems exactly like the type of cookie cutter tasks that the article describes would become like cashiers and have their wages stagnate. Meanwhile the studies with negative results were far more realistic and measured actual results. Otis et al 2023 measured profits and revenue of actual Kenyan SMBs. Roldan-Mones measured debate performance as judged by humans.

Re: AI is a floor raiser, not a ceiling raiser

#102

Post by jjk166 »

exasperaited wrote: post #45
throe23486 wrote: post #19

Earlier quoted context omitted.

I read this as interloper. What's an extraloper?

Opposite of "inter-" is "intra-". Intraloper, weirdly enough, is a word in use.

"inter-" means between, "intra-" means within, "extra-" means outside. "intra-" and "inter" aren't quite synonyms but they definitely aren't opposites of eachother.

Re: AI is a floor raiser, not a ceiling raiser

#104

Post by nomel »

LeftHandPath wrote: post #27

There are some things that you still can't do with LLMs. For example, if you tried to learn chess by having the LLM play against you, you'd quickly find that it isn't able to track a series of moves for very long (usually 5-10 turns; the longest I've seen it last was 18) before it starts making illegal choices. It also generally accepts invalid moves from your side, so you'll never be corrected if you're wrong about…

> you'd quickly find that it isn't able to track a series of moves for very long (usually 5-10 turns; the longest I've seen it last was 18)

In chess, previous moves are irrelevant, and LLM aren't good with filtering out irrelevant data [1]. For better performance, you should include only the relevant data in the context window: the current state of then board.

[1] https://news.ycombinator.com/item?id=44724238

Re: AI is a floor raiser, not a ceiling raiser

#105

Post by bloomca »

serial_dev wrote: post #77
cropher wrote: post #74

Really liked this article. I wonder: the graphs treat learning with and without AI as two different paths. But obviously people can switch between learning methods or abandon one of them. Then again, I wonder how many people go from learning about a topic using LLMs to then leaving them behind to continue the old school way. I think the early spoils of LLM usage could poison your motivation to engage with the topic o…

I learn about different subjects mixing traditional resources and AI. I can watch a video about the subject, when I want to go deeper, I go to LLMs, throw a bunch of questions at it, because thanks to the videos I now know what to ask. Then the AI responses tell me what I need to understand deeper, so I pick a book that addresses those subjects. Then as I read the book and I don’t understand something, or I have some…

I use a similar approach. I tried to experiment going into a topic with no knowledge and it kinda fumbles, I highly recommend to have an overview.

But once you know basics, LLMs are really good to deepen the knowledge, but using only them is quite challenging. But as a complementary tool I find them excellent.

Re: AI is a floor raiser, not a ceiling raiser

#106

Post by dsjoerg »

smiley1437 wrote: post #31
LeftHandPath wrote: post #27

There are some things that you still can't do with LLMs. For example, if you tried to learn chess by having the LLM play against you, you'd quickly find that it isn't able to track a series of moves for very long (usually 5-10 turns; the longest I've seen it last was 18) before it starts making illegal choices. It also generally accepts invalid moves from your side, so you'll never be corrected if you're wrong about…

> people aren't aware of how wrong they can be, and the errors take effort and knowledge to notice. I have friends who are highly educated professionals (PhDs, MDs) who just assume that AI\LLMs make no mistakes. They were shocked that it's possible for hallucinations to occur. I wonder if there's a halo effect where the perfect grammar, structure, and confidence of LLM output causes some users to assume expertise?

> I have friends who are highly educated professionals (PhDs, MDs) who just assume that AI\LLMs make no mistakes.

Highly educated professionals in my experience are often very bad at applied epistemology -- they have no idea what they do and don't know.

Re: AI is a floor raiser, not a ceiling raiser

#107

Post by righthand »

falcor84 wrote: post #33

I agree with most of TFA but not this: > This means cheaters will plateau at whatever level the AI can provide From my experience, the skill of using AI effectively is of treating the AI with a "growth mindset" rather than a "fixed" one. What I do is that I roleplay as the AI's manager, giving it a task, and as long as I know enough to tell whether its output is "good enough", I can lend it some of my metagcognition…

How do you know it’s significantly better quality if you don’t know any of the “how”? The quality increase seems relative to the garbage you start with. I guess as long as you impress yourself with the result it doesn’t matter if it’s not actually higher quality.

Re: AI is a floor raiser, not a ceiling raiser

#108

Post by LtWorf »

pdntspa wrote: post #66
erlend_sh wrote: post #10

Only for the people already affluent enough to afford the ever-more expensive subscriptions. Those most in need of a floor-raising don’t have the disposable income to take a bet on AI.

It's very easy to sign up for an API account and pay per-call, or even nothing. Free offerings out there are great (Gemini, OpenRouter...) and a few are even suitable for agentic development.

And how long until they raise the prices?

Re: AI is a floor raiser, not a ceiling raiser

#109

Post by LtWorf »

intended wrote: post #18
erlend_sh wrote: post #10

Only for the people already affluent enough to afford the ever-more expensive subscriptions. Those most in need of a floor-raising don’t have the disposable income to take a bet on AI.

Either you are the item being sold or you are paying for the service. Nothing is free, and I for one prefer a subscription model, if only as a change from the ad model. I am sure we will see the worst of all worlds, but for now, for this moment in history, subscription is better than ads. Let’s also never have ads in GenAi tools. The kind of invasive intent level influence these things can achieve, will make our curr…

Normally even if you pay you're still the product anyway. See buying smartphones for example… you pay a lot but you're the product.

Re: AI is a floor raiser, not a ceiling raiser

#110

Post by famouswaffles »

LeftHandPath wrote: post #27

There are some things that you still can't do with LLMs. For example, if you tried to learn chess by having the LLM play against you, you'd quickly find that it isn't able to track a series of moves for very long (usually 5-10 turns; the longest I've seen it last was 18) before it starts making illegal choices. It also generally accepts invalid moves from your side, so you'll never be corrected if you're wrong about…

LLMs playing chess isn't a big deal. You can train a model on chess games and it will play at a decent ELO and very rarely make illegal moves(i.e 99.8% legal move rate). There are a few such models around. I think post training messes with chess ability and Open ai et al just don't really care about that. But LLMs can play chess just fine.

[0] https://arxiv.org/pdf/2403.15498v2

[1] https://github.com/adamkarvonen/chess_gpt_eval

Post reply on HN