HNHacker News
TopNewBestAskShowJobs

mrtranscendence

1,366 karma · joined August 5, 2020

submissionscomments
mrtranscendence··on Microsoft, OpenAI sued for ChatGPT 'privacy violations'
> Your guess is that the majority of Reddit users have joined since 2018?

It's not really important to the debate around unlicensed use of copyrighted works to train AI models, but it wouldn't surprise me at all if the majority of Reddit users have joined since 2018. It's tough to get reliable active user counts, but they seem to have risen substantially over the past five years.

It also wouldn't surprise me if the majority of Reddit users were indeed from prior to 2018, but at the very least > 2018 would be a very substantial minority.

mrtranscendence··on Microsoft, OpenAI sued for ChatGPT 'privacy violations'
Well, you could memorize and recite copyrighted works all you want, as long as you're doing it in an empty room without anyone listening.
mrtranscendence··on Microsoft, OpenAI sued for ChatGPT 'privacy violations'
There are such datasets, and AI companies absolutely pay to have data curated. But I suspect it would be just unimaginably expensive to create a dataset from scratch with enough tokens to feed a model with hundreds of billions of parameters, all the while paying every participant fairly.
mrtranscendence··on Microsoft, OpenAI sued for ChatGPT 'privacy violations'
Since the way GPT "learns" is not materially similar to how a human learns, I don't see why this talking point is particularly relevant. Nothing stops the courts from distinguishing between an AI and a human with regard to what may be permissible.
mrtranscendence··on LLM Powered Autonomous Agents
I guess I'm just marginally more sanguine than you are that issues related to infrastructure and logistics could be, eventually, worked around. There would still be ways to contact people over a distance, some of which would remain available even in the initial chaos. Further, there are many people still working today who worked before internet-connected-everything, so there may still be useful institutional knowledge to bring to bear on these new problems.

My point on the iPhone was about what would happen after telephony infrastructure and supply chains had recovered (however long that takes). In the immediate aftermath of the shutdown mobile phones would absolutely lose the ability to make or receive calls. But again, I expect this to recover somewhat after a few months.

Your point on hospitals not being able to order supplies is a good one, regardless. Without blood supplies or dialysis equipment (or a bunch of other things), people would absolutely die. Perhaps those first weeks would be deadlier than I anticipated.

mrtranscendence··on Integrated Information Theory
Well, there's a reason it's called "the hard problem of consciousness" and not "just some incidental observation about models of consciousness". When evaluating general relativity it's not important to ask how mass/energy causes the curvature of spacetime; that's not the point of general relativity. But a theory of consciousness that doesn't explain the most intriguing, important component of consciousness doesn't explain what's important to me.
mrtranscendence··on LLM Powered Autonomous Agents
I think the consequences of a sudden, permanent shutdown of the internet would be fairly dire over the short term. Grocery stores would face inventory problems, with shortages worse than what we saw in 2020. Emergency services could be stretched thin as they deal with any unrest, while navigating their work without any internet-connected services they relied on. Energy infrastructure could experience interruptions due to the loss of connectivity.

As for economic consequences, millions of jobs would just no longer make sense, and many others would get harder and frequently much less efficient. Certain major categories of product no longer make as much sense; what's an iPhone that can't connect to the internet worth? It's just a telephone with a camera (that takes images you can't share).

It could be deadly to some degree? All it takes is for the power to go out somewhere exceedingly hot in July, as the loss of air conditioning can be deadly for the elderly. But as deadly as the nukes the US dropped on Japan? I don't see it.

mrtranscendence··on LLM Powered Autonomous Agents
I haven't studied diffusion methods in great detail, but my simplistic ELI5 understanding goes something like: during training you gradually add noise to an image, step by step, and then the model learns how to remove the noise at each step. Adding noise to an image is, of course, very easy to do.

But for this approach adding "noise" would be much less straightforward, wouldn't it? You'd have to have some way to work out how to take "Berlin is the capital city of Germany with ..." down to an empty string in a way that marginally removes detail. For some passages this could be rather difficult to do. I feel like gathering training data would be a huge trial here.

mrtranscendence··on Integrated Information Theory
I rarely see IIT characterized as complete junk. Personally, I don't see much value in it as I don't think it answers, or even grapples with, the hard problem of consciousness. Tell me how integrated information gives rise to subjective experience and maybe I'll start buying what they're selling.
mrtranscendence··on Integrated Information Theory
I don't think IIT is a correct theory or even on the right track. Nevertheless I fail to see how it's "silly". It's no sillier than any other putative scientific theory of consciousness -- less silly, since IIT actually makes a testable prediction.
mrtranscendence··on Framework Laptop 16 Deep Dive – Enclosure
I'm not sure I follow you here. I'm not saying that it should necessarily come with RAM, storage, and OS, which can be removed without disassembling the entire laptop. And I'm not making a statement one way or the other on whether you'll have to ensure your computer works.

I'm only saying that it's funny to learn that Framework undergoes some effort to assemble and test the laptop only to undo that work so the buyer can assemble the entire thing. It might be for entirely good reasons, but that doesn't make it a bit funny prima facie.

mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
I do the same thing, though weirdly enough less than I used to. Sometimes it really helps me work through a problem -- maybe it slows me down to a better pace? Who knows.

Talking to pets also works. They've never been able to offer a very useful response, alas.

mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
I do tend to talk through spatial problems, yes, though I'm not so aphantasiac that I can't visualize, say, the image that would match a simple pattern of lines demonstrated in prior images. I can also situate things with relative position in my head without having to visualize them (if that makes sense).

> If you see an art piece, do you only appreciate it through the monologue, or does it feel beautiful in a non-verbal way too?

I appreciate it both ways, yeah. But the visual in that case would mostly be just "this is pretty" or "that's disturbing" etc. Any kind of genuine art critique (such as I am even capable of doing) would be talked through mentally.

mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
> By default, while I read something there's no auditory element at all.

I wouldn't characterize how I read -- which as far as I know is normal, but who can tell! -- as narrating to myself. Rather, it involves the sensation of somehow pronouncing words without necessarily hearing them. I can do this much faster than I could narrate the text explicitly.

I can also read without either narrating or pronouncing, but I find that it cuts down comprehension noticeably.

mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
I've gone back and forth on aphantasia. It's true that I generally don't see images in my mind; even when thinking about things that are visual I'll tend to mentally describe them in English. But if I really try I can definitely see things in my mind, after a fashion, although it's very ... blurry, for lack of a better word.

I'm also terrible with faces. I can see someone I've known for years with a different haircut and I'll completely fail to recognize them. If everyone suddenly went bald I'd be helpless to recognize anybody.

Edit: Like, I'm right now trying to conjure up the face of my girlfriend of 23 years (who is on a business trip right now). I just can't do it. It's kind of trippy.

mrtranscendence··on Framework Laptop 16 Deep Dive – Enclosure
Apropos of nothing, I learned (thanks to a recent Linus Tech Tips video) that Framework's DIY Edition laptops are preassembled for QA and then disassembled prior to shipping. I found that amusing. Not sure I could buy a DIY model knowing that.
mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
You don't need to be rich to buy them, but you do have to care at least a little bit to spend $170 on headphones (or $250 for the Pro). I think they've continued to be a status symbol to some degree.

I could see a kid wanting AirPods because his cool friend has them. I don't think I'd see the same thing with Skullcandy, or Bose for that matter.

mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
I've spent at least $3500 on my gaming PC, and I may even get an AVP, if the reviews are extremely good. But I would never fly business class unless my company were paying for it (which they only will for international flights ... and we're strictly US-only). It's absurdly expensive and the delta over coach is minimal.
mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
I've seen one used at work for R&D demo purposes, but aside from that I don't think I've ever noticed anyone using one while out and about (Cincinnati, OH). Certainly never in transit or on an airplane.
mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
In the US the initial iPhone was a bit of a sensation. Demand outstripped supply at first, and people waited outside stores for hours to snag one. It sold less than future models would, because it seemed fairly expensive and the value proposition wasn't yet 100% clear to many. But basically everyone knew about the iPhone and had an opinion one way or the other, generally positive.

Even now the iPhone isn't as popular in mainland Europe as it is in the UK, Ireland, and the US.

mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
I don't hate it, but I'll never be anywhere near as fast as on a real keyboard. It's a shame there aren't new paradigms for input. I loved the Nintype keyboard, which combined swiping with tapping for impressive speeds (though still less than a real keyboard), but that's abandonware at this point.
mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
Are you aware how successful those products have been? They don't match iPhone revenue by any means, but they move a ton of units and are the market leaders for their categories.
mrtranscendence··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
> Corporations are made up from people and many of those who developed their most successful products initially are no longer there

True, but it's not as if this is a completely new company with new employees. Much of the leadership remains, many of the rank and file as well, and probably many internal processes/guidelines. I'm comfortable claiming that Apple's past successes are relevant when considering whether the AVP will be successful as well.

Which isn't to say it will be successful. The price alone will ensure it remains niche. But apart from cost I fully expect it to be a better headset in many ways than others that came before it.

mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
That seems insane to me. I'm always chattering away up there, I can't imagine it otherwise.
mrtranscendence··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
There may be upsides to the ability to fake video and audio of someone (better CGI effects in films, for example). But in my experience when people refer to a deepfake they seem to mean that the fake has been distributed to confuse or deceive, for which I can't really see any probable upsides.

> So basically people and civilization are going to muddle along as they always have.

I agree with this, but consider the drawbacks to rampant disinformation and the proliferation of deepfakes (all this is IMO): it will make any video or audio deniable and unusable as evidence. Real images will be denounced as fakes. Fake images will catch on and possibly cause real damage. People will rapidly lose trust in most sources of news, entrenching established known quantities.

I feel like if we could reasonably put a stop to this we should. I don't think we can in general, though.

mrtranscendence··on GPT4 is 8 x 220B params = 1.7T params
I think the question stands. You can't fit 25 4090s in a robot (unless we're talking about something massive with an equally massive battery), and even if you could an LLM wouldn't be appropriate for driving a robot.

Given the pace of improvements I don't see how you compress 25 4090s into a single GPU in 7 years. A 4090 isn't 25 times the power of a GTX 980, it's closer to maybe three times.

mrtranscendence··on GPT4 is 8 x 220B params = 1.7T params
This doesn't seem likely in the near term. An iPhone 13 Pro has 6gb of memory, which might be enough for a single 1.3b parameter model if you purged everything else out of RAM. Combining 8 (or even 4) of them on a single phone won't happen anytime soon at the rate phones improve.

Plus, none of the smaller models are really appreciably close to GPT-4 on most metrics. It's not clear to me you could get there at all with 1.3b models. Maybe somebody gets there someday with 65b models, but then you're far out of the reach of phones.

mrtranscendence··on GPT4 is 8 x 220B params = 1.7T params
1.7 trillion patients? We're going to need more hospital beds ...
mrtranscendence··on Ask HN: Has Google search switched to infinite scrolling for you too now?
I’m so hopelessly addicted to YouTube that premium only made sense. I should probably work on that, though.
mrtranscendence··on Deep Learning’s Diminishing Returns (2021)
For me, the takeaway is that training state of the art models may become cost prohibitive even for large companies. There may be a practical limit to how complex models can become. This is at least plausibly interesting to know.
← PreviousPage 4 of 32Next →