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

There is no hard takeoff

geohot.github.io

Post reply on HN
Print view View original

161–170 of 190 posts

Re: There is no hard takeoff

#161

Post by pusspuss »

Eliezer persistently croons dark lullabies of AGI cataclysms. While his obsession is evident, one cannot help but wonder if these incessant dirges might be nothing more than echoes of groundless fears. Much like the puppet that cries out every time the stage darkens, only to find there's no true end approaching, we may find ourselves dulled to genuine threats, leading to our eventual and inescapable undoing.

Moreover, by focusing too much on distant threats, we may overlook current problems, like companies gaining too much control through AGI. We don't need AI safety rules that give big players more power and shut down open-source projects.

Re: There is no hard takeoff

#162

Post by isodnhionio »

leshow wrote: post #159
isodnhionio wrote: post #133

Earlier quoted context omitted.

This is yet another reason I want Musk in prison. Getting drivers off the road will save lives, and not a small number. Most companies investing in automated driving have built a system that is safe and largely effective. These systems have worked safely in the places they've been deployed. We should be heavily investing in expanding their capabilities. Yet Tesla's system is the most famous and most widely-deployed a…

It seems to me if you wanted to deliver on "economic benefits", save lives, reduce emissions, etc there is an obvious already existing option: subways/LRT. For inter-city travel build high speed rail. I just don't see driverless cars as the panacea that other people do, even if it worked— which it doesn't seem to.

Nothing has ever been a panacea, and people very rarely claim things will be. Cars will be around for a long time. Building public transportation is not trivial and not practical everywhere. People often drive even in parts of the world with good public transportation. Replacing human drivers with (working) computers would be a major improvement.

Re: There is no hard takeoff

#163

Post by KnobbleMcKnees »

fisf wrote: post #158
KnobbleMcKnees wrote: post #152

Earlier quoted context omitted.

The implication was that his technical knowledge is very good and his writing is not, but you missed that while you were getting angry.

That's not really what "puddle deep knowledge of anything that isn't iPhone firmware" implies.

Again you are free to read it how you wish, biases and all.

Re: There is no hard takeoff

#164

Post by unsubstantiated »

scaramanga wrote: post #40

Am I the only one who sees a jumbled stream of barely coherent semi-thoughts and word salad?

> Am I the only one who sees a jumbled stream of barely coherent semi-thoughts and word salad?

Probably one of the few given the majority of the comments are directly addressing and dissecting each point of the post.

It is not jumbled, but it is 1) the way Hotz usually writes, 2) unconventional, 3) cross-domain laterally connected, and 4) somewhat heavy for a sunday morning read.

Hotz generally doesn't do "barely coherent semi-thoughts." I think the label is needlessly harsh. He's just addressing the contemporary nonsense by bridging them together and calling their bullshit out.

It's sad to see so many of the replies to your comment are people attacking him for no good reason.

Re: There is no hard takeoff

#165

Post by gitfan86 »

optimalsolver wrote: post #52
reedf1 wrote: post #32

He makes a good point. The only real way to "summon the demon" is if AGI is centralized. If each of us has a maximally intelligent AGI in our phone - and it is us as individuals controlling the prompts, then the playing field is levelled. Unstoppable force meets unstoppable force. Gun meets gun. A tale as old as society.

Would every citizen having their own personal nukes make the world a more pleasant place to live?

Can a nuke perform manual labor jobs and drive cars 100x safer?

I don't think so, so it isn't a good comparison.

A better comparison would be guns.

There is a valid debate to be had on gun control

Re: There is no hard takeoff

#166

Post by benlivengood »

The argument falls apart when considering that compute is incredibly fungible and performing a 51% attack only means finding a zero-day that lets an AGI copy its code or neural net to a large percentage of computers. Much like wealth accumulates in the absence of antitrust law and taxation, intelligence accumulates in the absence of some external controls.

EDIT: And probably all you need is natural selection. No one is running GPT-2 anymore except for basic research. Any AGI with marginal advantage over others will be run instead of them. At some level of complexity identical agents will coordinate as one entity with shared goals.

Re: There is no hard takeoff

#167

Post by georgehotz »

tysam_and wrote: post #64

I think this is a good casual introduction to the marketplace dynamics of how ML will impact the market. I do, however, disagree as I feel that this version of things is a bit too sterile, theoretical, and 'academic', and assumes a more open-information set of competitive strategies among potentially ideal agents from a game theoretic perspective. We can see this is absolutely not the case 'in real life'. To blatantl…

lol, you should see me bash my own code. I'm even more mean.

https://github.com/tinygrad/tinygrad/blob/master/examples/hl...

have a bunch of bounties on it, we're getting 94%+ now! mostly not me who wrote this, see history. have to switch to float16 and add Winograd convs still. we have a branch with multigpu too.

goal is to beat an A100 in speed on a tinybox.

Re: There is no hard takeoff

#168

Post by jareklupinski »

micheljansen wrote: post #34

This article reminds me a bit about a core theme in Isaac Asimov's Foundation series, which I'm coincidentally reading (mild spoiler warning). The science of psychohistory that is used to predict the future and create the Seldon Plan, only works if the future is allowed to unfold uninformed by psychohistory itself. That's why the Foundation was founded without any psychologists or knowledge of the Seldon plan. Seldon…

> knowledge of those predictions changes the outcomes, so they are no longer reliable

does the series explore whether the outcomes change for the better, for the worse, or simply for the different? curious to read them, I'm thinking about writing something similar, but with positive consequences for future knowledge

Re: There is no hard takeoff

#169

Post by latchkey »

> Nobody has a 1e20x more efficient Bitcoin miner either

The miners today are 1e20x more efficient than the original CPU based miners. I mean, maybe not exactly 1e20x, but we saw insanely massive improvements, in only 9 years.

> Now, if there’s one weird trick to 1e20x your efficiency, and only one group gets it, all bets are off.

Only one group really did get the massive efficiency increases... Bitmain, a Chinese company. They sold machines and even mined with them before shipping them to customers.

Are all bets off with Bitcoin today? Nope.

Re: There is no hard takeoff

#170

Post by goosinmouse »

threeseed wrote: post #30
mysterydip wrote: post #20

Do we have enough historical weather data to train a decent AI on? Maybe this is already being done and is just a quieter area of research (in the headline sense)? I know for example every time a hurricane comes toward land all the weather places trot out a half dozen different model predictions saying different things. It would be great if there was more confidence in this area for planning.

> the weather places trot out a half dozen different model predictions saying different things This should give you a hint that it's not as simple as having enough weather data. You are trying to predict a chaotic weather system that is impacted by everything from sea temperatures, sun activity, wind, volcanoes, bushfires, atmosphere changes etc. Much of which we don't even have enough data for or a complete understa…

Yep, the main problem is the bottleneck of data. I mean think about it, we still fly planes full of scientists directly into the largest storms ever created, just to get a few points of data. Satellites can only do so much.
Post reply on HN