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

The Intelligence Age

ia.samaltman.com

Post reply on HN
Print view View original

251–260 of 447 posts

Re: The Intelligence Age

#251

Post by marcosdumay »

jay_kyburz wrote: post #236
whamlastxmas wrote: post #211

Earlier quoted context omitted.

I would wonder if solar plus the absolute massive battery reserves you’d need for a data center ten times bigger than the current worlds largest would still be cost effective versus a single fusion reactor

I wonder how that cost would compare to putting your solar and data center in space? (with non stop solar pointed at the sun) I wonder if cooling is easier or harder in space?

> I wonder if cooling is easier or harder in space?

It's harder. You can stop wondering.

It gets easier if you run things hotter, up to the point that there isn't much of a difference at a few hundred °C (and keeps making no difference when hotter).

Re: The Intelligence Age

#252

Post by benlivengood »

d_burfoot wrote: post #243

OAI's achievements are amazing. But here's a bit of a skeptical take: cheap human-style intelligence won't have a huge impact because such intelligence isn't really a bottleneck today. It's a cliche that the brightest minds of the age are dedicated to selling more ads or shuffling stock ownership around at high velocity. Anyone who's worked at a big tech company knows the enormous ratio of talent to real engineering…

Capital follows the most profitable investments. It's why "green" technology took so long to develop (we probably could have had efficient solar panels in the 60s, but oil was super cheap and more importantly very profitable). Dropping the cost of a $1e7 problem to even $1e6 probably makes it very profitable.

Re: The Intelligence Age

#253

Post by pdonis »

All this and not a word about AI ethics. If the ethics of the AI's Sam Altman is building are his ethics, I wouldn't trust them to do any of the tasks he describes, or indeed any tasks requiring any kind of independent action or judgment at all.

Re: The Intelligence Age

#254

Post by danielmarkbruce »

lxgr wrote: post #241
danielmarkbruce wrote: post #204

Earlier quoted context omitted.

Maybe don't think of it as curves or functions. Just go through an LLM and think about all the things that could be improved. It's a long list once you get into the details. By and large...sota models are: a) trained on crappy data, including questionable RLHF feedback. b) trained with questionable embedding layers. c) trained with questionable loss functions d) trained with questionable optimizers e) trained at ques…

I don't doubt that there's tons of work ahead to even just integrate current-day capabilities of LLMs into our society and economy. But when talking about future growth potential, I don't think you can get around making assumptions about the shape of the growth function.

You can't write a nice article about it. But you can talk about it with a simple "things are going to get a loooot better". The problem with thinking about progress in functions is it suggests some underlying law exists, and there isn't one.

Even if someone could point to a function and say "10x better" by 2030 - what does that even mean in the context of an LLM for example?

Re: The Intelligence Age

#256

Post by hackable_sand »

visarga wrote: post #95
joshmarlow wrote: post #85

Earlier quoted context omitted.

This dovetails with something I've been thinking lately; the foundation of civilization is standing on the shoulders of giants (ie, I might make small contributions but I didn't invent compilers, CPUs, mining, agriculture, etc). Progress in large part is figuring out better ways of doing that (language, written language, printing press, internet access, etc). When you look at things that way - LLMs start to seem deep…

> the foundation of civilization is standing on the shoulders of giants. Progress in large part is figuring out better way of doing that It's so much easier to imitate than to invent something truly novel and useful. Let's do a bit of napkin math. A human lifetime is about 500M words. GPT-4 used up about 30,000 human lifetimes of language. But cultural evolution took 200K years and 120B people to get here, about 4 mi…

Well put.

Re: The Intelligence Age

#257

Post by saati »

falcor84 wrote: post #244
oculty wrote: post #219

Earlier quoted context omitted.

That sounds really useful. Can you give an example of what kind of content your are uploading and how the tutoring looks like?

One example is of studying robotics. I started with this course on Coursera [0]. It was ok at the start, but got hard for me early on. A big part of the course is of reading the book that they make available for free [1], so I would upload the relevant chunk of chapters into Claude (the whole book was a bit too much for it), and would then just ask it to explain each topic to me, whereby I'd ask additional questions,…

How do you know the answers are actually useful and not just hallucinations?

Re: The Intelligence Age

#258

Post by impossiblefork »

codingwagie wrote: post #36
schmidtleonard wrote: post #35

Earlier quoted context omitted.

Easy for him to say: AI is almost guaranteed to hand a massive W to capital and L to labor. He is holding a title to rule over hell in one hand and promising to lead us to heaven with the other.

Actually not sure about this, with the leverage of AI, its easier than ever to start a company

Yes, so there will be all sorts of attempts, and the productive forces will compete with each other, while landowners and owners of harder to disrupt industries will have the power to choose among them, deciding who will be allowed to be successful and who will fail.

There isn't a single ordinary person at the table, no labour unions, no political parties or anything democratic. It's Microsoft, it's Google, it's some venture-capital owned French firm, some venture-capital owned German firm, etc. Maybe if Schmidhuber's stuff is good as he hopes it is maybe there'll be an Austrian firm, etc., but mostly, it'll be a capital intensive business controlled by people with capital.

Re: The Intelligence Age

#259

Post by marcosdumay »

ctoth wrote: post #119
swagasaurus-rex wrote: post #105

Earlier quoted context omitted.

While I agree these LLMs are getting pretty good, I think google shares fault for making it so hard to find information on the internet. You have to scroll past 4 sponsored links to get to their algorithm results, which themselves have been gamed by SEO and content farms. As far as I can tell, google has no plans to limit AI generated botfarm search results. LLMs now seem to get you an answer more quickly than google…

I don't really understand what this comment about Google has to do with OP's comment. Is this something that even old Google could have done pre-SEOification? Would Google have been able to read the datasheet? Or just point OP at what they already said they were happy to avoid? > Without AI this would have been a long long process of brushing up on probability, vector math, manually transcribing the chart. Sometimes,…

Are you asking if Google could ever find documentation and sample code on the internet?

Or are you asking if Google could understand a badly written query and understand it's about the sample code?

Because the answer to the first is "yes", and to the second is "no". But the first one adds almost all of the productivity here.

Re: The Intelligence Age

#260

Post by layer8 »

dartos wrote: post #175
layer8 wrote: post #65

Earlier quoted context omitted.

We also personify synapses and axons in human brain tissue, though. My point is, while I agree with your first sentence to a degree, we shouldn’t judge the whole solely by its elementary parts. Clearly an LLM exhibits very different behavior from a conventional database.

LLMs exhibit very similar behavior to a search algorithm. Text query in -> relevant text out. I don’t say that search algorithms “learn” or “think” outside of ML.

The algorithm that learns is the training algorithm whose output is the LLM, not the inference algorithm employed when using the finished model.
Post reply on HN