diff options
Diffstat (limited to 'blog/posts/ai2')
| -rw-r--r-- | blog/posts/ai2/index.md | 16 |
1 files changed, 8 insertions, 8 deletions
diff --git a/blog/posts/ai2/index.md b/blog/posts/ai2/index.md index 750d442..9769395 100644 --- a/blog/posts/ai2/index.md +++ b/blog/posts/ai2/index.md @@ -1,6 +1,6 @@ It's no secret that I'm generally not very fond of the Silicon Valley tech hype cycle we currently find ourselves in — in fact I've [posted](/blog/posts/ai_art/) about my feelings on "AI" technology before, and I think that post still mostly holds up, but having had circa 15 months since to ruminate about it and watch this shitshow pan out, my opinions on the matter have gotten substantially more nuanced, and I'd like to formally express them, ideally in a more level-headed way than my first go, so here's that. -## The still-bad +# The still-bad The primary product being inescapably pushed on us, generative AI (oft shortened to GenAI, and henceforth referred to here as genai<sup id="n1r"><a href="#n1">1</a></sup>), remains pretty self-evidently worthless, surprising nobody less than me. @@ -15,13 +15,13 @@ For one, the output quality of LLM chatbots and such things in particular [doesn </figcaption> </figure> -### Ethics? What? I think we fired those guys +## Ethics? What? I think we fired those guys Ethical problems also continue to abound. I'm not personally aware that [labeling practices](https://www.theguardian.com/technology/article/2024/jul/06/mercy-anita-african-workers-ai-artificial-intelligence-exploitation-feeding-machine) have gotten any better, and on top of that, poorly written, badly behaved scrapers run by popular genai vendors are now [effectively](https://www.gamedeveloper.com/business/-this-was-essentially-a-two-week-long-ddos-attack-game-ui-database-slowdown-caused-by-openai-scraping) [DDOSing](https://techcrunch.com/2025/01/10/how-openais-bot-crushed-this-seven-person-companys-web-site-like-a-ddos-attack/) [random websites](https://social.kernel.org/notice/AqJkUigsjad3gQc664) in their attempts to suck down as much data as possible for training. Many website admins, especially those running git forges, which are targeted by coding assistants like GitHub Copilot, have had to resort to countermeasures like [Anubis](https://anubis.techaro.lol/) to protect their sites from this behaviour. I've had a run-in with this activity personally, when a big spender on Alibaba Cloud slammed my [Nepenthes](https://zadzmo.org/code/nepenthes/) instance so hard I had to [set up a custom Fail2ban jail](/blog/posts/grow_spikes/) that blocked over 300 IPs inside a day just to make its statistics endpoints reachable again. Furthermore, the environmental impact of this technology [remains really bad](https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/). Electricity use alone has gotten so bad that, for example, Microsoft has had to [spin a decommissioned nuclear power plant back up](https://www.npr.org/2024/09/20/nx-s1-5120581/three-mile-island-nuclear-power-plant-microsoft-ai) to run their datacentres. Although, some amount of proliferation and normalization of nuclear energy would be a nice little upside to this whole debacle. -## The actually good +# The actually good The class of algorithm that "AI" as a marketing term refers to, that is to say, machine learning and neural networks, actually has some compelling use cases that we've seen crystallize over the past couple years, and I think they merit mention here. @@ -31,7 +31,7 @@ Genai is also kind of just the best way to make good-sounding text-to-speech and Of course, tech like this isn't the *point* of the hype cycle. Nvidia didn't become the first trillion dollar company by inventing a niche technology to make new GeForce video cards perform better in demanding games, and Microsoft isn't desperate to make "AI PCs" a thing because of the potential for better screen readers. The "AI" hype cycle is about genai supposedly having the potential to revolutionize *everything*. In five more years. And please don't ask how. -## The it's-complicated +# The it's-complicated LLMs are, by nature, endless fountains of grammatically-fluent drivel utterly devoid of real meaning. This makes it anywhere from unsettling to infuriating to see them used in contexts such as education, journalism, or software development<sup id="n3r"><a href="#n3">3</a></sup>, but it makes them uniquely suited to two use cases: spam, and the sort of high-verbosity, low-meaning communication that's expected of you in a professional setting — in other words, the LLM has democratized the least valuable forms of written communication in existence. @@ -43,7 +43,7 @@ This is a pattern I've noticed with a lot of the more infamous uses of genai pro Even if completely ridding the world of genai technology was possible at this point, it wouldn't be the solution to these sorts of problems. The solution is to destroy the incentive structures that motivate this behaviour in the first place — create a school system where students don't feel the need to cheat on schoolwork, remove the profit incentive to generate spam and replace human workers with cheap, unreliable machines, etc. In other words, the real problem is capitalism. -### The AI art thing +## The AI art thing Then there's the issue of "AI art". I stand by the it's-not-art-on-a-technicality argument I made before — I still don't think typing a prompt into Stable Diffusion and then taking the output and pretending you "made" something constitutes any kind of real artistic contribution, and frankly the only worthwhile artistic statements I've seen made using these tools have been statements *about* genai. I also tend to find the aesthetics of the output of genai thoroughly repulsive in general. The silver lining here is that blatant use of genai to make, say, game assets or promotional art is usually a pretty damn good indicator that whatever the slop is being poured on was never worth your time to begin with. @@ -60,7 +60,7 @@ Let's get one thing straight, however: If you use genai to generate memes or, li </figcaption> </figure> -## They did a conclusion on this one +# They did a conclusion on this one Here's hoping this one comes off as less of an unhinged rant than the last time I posted about this. This is by no means a complete compendium of every thought I've had about genai, but it is everything I think I have something substantial to say about. For example, the whole concept of deepfakes is pretty horrifying, but I don't have anything to add to that conversation beyond gesturing vaguely at it and going like "wow, that's pretty fucked up, huh? anyway im rod serling". @@ -68,7 +68,7 @@ It's clear at this point that genai is here to stay in at least *some* capacity, --- -## Further reading +# Further reading Linked above: @@ -93,7 +93,7 @@ Linked above: <p id="n1"><a href="#n1r">1</a>. Pronounced "gen-eye", with the "G" sound from "graphics". This is to annoy people who think genai is cool.</p> -<p id="n2"><a href="#n2r">2</a>. In particular, Microsoft's insistence on this garbage is <span style="color: yellow;">Pissing</span> me off. I'm genuinely considering (<strong>update</strong>: I did) getting a Macbook for some upcoming Adobe-requiring classes just to avoid Copilot/Recall/whatever nonsense they come up with next.</p> +<p id="n2"><a href="#n2r">2</a>. In particular, Microsoft's insistence on this garbage is <span style="color: yellow; padding: .2em .5em; background-color: black; border: 1px solid white;">Pissing</span> me off. I'm genuinely considering (<strong>update</strong>: I did) getting a Macbook for some upcoming Adobe-requiring classes just to avoid Copilot/Recall/whatever nonsense they come up with next.</p> <p id="n3"><a href="#n3r">3</a>. If you unironically self-identify as a "vibe coder" I'm coming to your house and confiscating every internet-capable device you own. Hand over the MacBook, dipshit, you've lost your computer priveleges until you're ready to develop a working understanding of how the technology you're working with professionally actually functions.</p> |