TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Is Stable Diffusion a big deal?

27 点作者 jamiegreen超过 2 年前

5 条评论

Varyag超过 2 年前
These technologies are going to have a deleterious effect on the work available for artists, I think that much can be safely asserted. Of course there&#x27;s no way of really knowing what percentage of artist work this is going to eat up, but neverytheless: When coal miners lost their jobs to automation, they were told that the advancement of technology will create new, better, safer and more fulfilling jobs for people. But what happens when the fulfilling, enjoyable jobs get automated?<p>We live in a system that ties productivity to human worth, that (in the US, certainly) only gives basic human needs such as food, shelter and healthcare to those that are making money for the system. What happens when larger and larger percentages of the population are forced to choose butween uninspired, miserable work and a life of poverty and desperation? What happens when the productivity is locked to those with the wealth to purchase expensive automation tools to do the work, and the wealth circulates between the people making the automation and the people using the automation, with vast swathes of the population being excluded from opportunity?<p>I&#x27;m excited to see this technology move forward, but I&#x27;m more excited about what societal effects this type of automation will have.
评论 #32847129 未加载
marstall超过 2 年前
missing in all of this: &quot;the new&quot;.<p>Newness is what gives art, writing, news articles, discourse in general, its value. stable diffusion, GPT-3, etc are simply generating aimless rehashes of what&#x27;s already been said and done.<p>example: check out a daily newspaper, online or off. most every sentence of its news articles are based on new facts personally witnessed by the writer in the past week, expressed with the maximum of economy. where&#x27;s the room for an AI there?<p>Or fine contemporary art (the kind people pay for). This category of art inevitably incorporates things like <i>commentary</i> on previous generations of art, <i>rejection</i> of previous forms, <i>questioning</i> what an art form is, etc. None of these things can be done by Stable Diffusion&#x27;s pastiche of genres and influences.<p>Quickly artists will begin incorporating Stable Diffusion and its implications into their work - but they will create value only by doing things like <i>commenting</i> on it, <i>questioning</i> it, <i>perverting</i> it and <i>rejecting</i> it. And doing so will quickly become passé after it has generated some eye popping art sales.<p>Of course nothing ever generated by Stable Diffusion will have a dollar value on its own, what it outputs is the very definition of a worthless commodity.<p>For an earlier example of this phenomenon, check out the work of 80s artist Mark Kostabi, who created his own Stable Diffusion-esque entity in the form of a studio of hack artists who created an endless conveyor belt of oil paintings based on his whimsical, banal specifications. They were terrible and Kostabi was variously viewed as a genius and a laughingstock.
评论 #32835397 未加载
评论 #32841947 未加载
评论 #32833509 未加载
评论 #32835864 未加载
mturmon超过 2 年前
The casually-made claim:<p>&gt; &quot;How can a human designer compete?&quot;<p>and the (related, but as noted by @marc_io nearby, quite different) claim:<p>&gt; &quot;inevitable that this would impact on the demand for human artists, and I think this will start soon&quot;<p>both need a more careful look.<p>Regarding the &quot;designer&quot; part -- most people aren&#x27;t really very good at design and illustration, and adding computers doesn&#x27;t always help. See, GeoCities home pages, or your favorite desktop publishing real estate agent ad.<p>People do get tired of sameness in images, and their eyes like new things, personal styles, clarity, chaos, whatever. That&#x27;s one of the things designers do. See the <i>Helvetica</i> documentary for more [<a href="https:&#x2F;&#x2F;www.hustwit.com&#x2F;helvetica" rel="nofollow">https:&#x2F;&#x2F;www.hustwit.com&#x2F;helvetica</a>].<p>Point of comparison: Geoff Hinton&#x27;s careless claim, made in 2016, that radiologists would be displaced by 2021.<p>[<a href="https:&#x2F;&#x2F;towardsdatascience.com&#x2F;why-ai-will-not-replace-radiologists-c7736f2c7d80" rel="nofollow">https:&#x2F;&#x2F;towardsdatascience.com&#x2F;why-ai-will-not-replace-radio...</a>]<p>[<a href="https:&#x2F;&#x2F;statmodeling.stat.columbia.edu&#x2F;2021&#x2F;06&#x2F;07&#x2F;ai-promised-to-revolutionize-radiology-but-so-far-its-failing&#x2F;" rel="nofollow">https:&#x2F;&#x2F;statmodeling.stat.columbia.edu&#x2F;2021&#x2F;06&#x2F;07&#x2F;ai-promise...</a>]<p>This tech will have an effect, but thinking that design (or, breathtakingly, &quot;art&quot;), is now a solved problem and we can move on, seems premature.<p>A re-watch of the Helvetica documentary might help here. Desktop publishing changed print design but high-impact print (or 2D) design is still very important and not left to amateurs, for good reasons.
评论 #32840659 未加载
marc_io超过 2 年前
The author uses the terms “designer” and “artist” interchangeably. These are not the same activities at all.
评论 #32826178 未加载
aaroninsf超过 2 年前
IMO the big deal is not the small thing that is driving &quot;debate&quot; and backlash. The small thing is, we have now more or less solved the automation of the idea to image pipeline.<p>The big deal, or rather deals, are still emerging; and they are big because they are going to be far more disruptive and transformational than the relatively minor crises for working artists and designers that are relatively small. Relatively.<p>The big things are the things that you can see if you pay attention to the rough edge of what people are experimenting with and doing with these tools, beyond simply using them as drop-in replacements for talented artists&#x2F;designers.<p>The first thing people do with new tools is what they are often conceived of and intended to be used for by their makers: drop-in replacements for existing tools, but &quot;better&quot; for some values of better. Faster, cheaper, more reliable, etc.<p>The <i>interesting</i> things that people do with new tools, are all the new things that can <i>only be done with those with those tools</i>. There is a multidimensional territory that StableDiffusion and MidJourney etc are opening up, and it&#x27;s not just about media objects proper. It&#x27;s <i>also</i> about renegotiations of our relationship to various classes of media object.<p>The word &quot;flood&quot; is used to decry the inundation of generated content. The real story though is that the cost, and <i>half-life</i>, of images of &lt;whatever&gt; quality is rapidly going to zero.<p>We are watching in real time as striking, thought-provoking, captivating imagery [and beyond], is becoming ephemeral and on-demand and tailored to an audience of one.<p>Another longer-term but even bigger non-linear projection of where these systems are going, is about the <i>role</i> they occupy.<p>In the last few days the stratechery.com article on AI &quot;unbundling&quot; has been making rounds.<p>The analysis is basically sound. But it falls short of identifying the <i>big</i> story, that the step by step automation of the entire &quot;content generation&quot; chain has not reached it its end.<p>Now, we are in the last moment when humans are necessary as collaborators at all. We still provide executive functions: intention, discrimination, filtering, ideation.<p>But automation of those things wrt to the media stream and cultural discourse (aka the Zeitgeist) which individuals consume permute and articulate their &quot;own&quot; ideas about, is quite obviously not just inevitable, but going to be done &quot;better&quot; for various values of better, by these seems types of systems.<p>Will that be &quot;soon&quot;? Well, no; but it&#x27;s &quot;a simple matter of engineering&quot;—and it will happen.<p>So... the <i>big</i> story here is that in the near (sic) future, we will live in a world in which superhuman art (and design) is generated for us on the fly. Influenced by and steered, sort of, when we try, to our tastes—but mostly devised for us, on the basis of superhuman discrimination...<p>...and with intentions and aims we better get a hold of.<p>The Alignment Problem here is going to redefine society.<p>This is what I mean: consider for a moment the surveillance society we live in today; and what that surveillance is for: to understand and steer our behavior, for venal economic purposes—and venal political ones.<p>Now extrapolate to a world in which the discernment, and <i>steering</i>, are effected by superhuman tools.<p>That&#x27;s the biggest deal IMO.
评论 #32828725 未加载
评论 #32828686 未加载