Stephen Farrugia

RE: https://hci.social/@fasterandworse/116586227741468965

You can't tell the world you use AI for *just this small thing* without contributing to the legitimacy of the whole AI industry.

You also can't use AI for *just this small thing* and think that you won't become the next *just this small thing*.

May 19 at 5:30pmWeb
Show threadStephen FarrugiaMay 19
I won't try to argue the ethical issues of using LLMs — because, as real as they are, when you use it for *just this small thing* you've made your choice.
Show threadStephen FarrugiaMay 19
I don't doubt how useful you believe it is for *just this small thing*, but I do doubt how much you really care about *just this small thing* to accept a product which is incidentally useful instead of recognising that any legitimate purpose deserves a product legitimately designed to satisfy it.
Show threadStephen FarrugiaMay 19

LLMs are not designed for a purpose. They are not designed for a group of related purposes. They are not a product of design.

They are a mechanism of design which is being reactively presented as solutions for whatever purposes seem to have the most revenue potential.

Show threadStephen FarrugiaMay 19

Any product which is not designed for an *anchoring* purpose is doomed to chase revenue how ever it can.

Your *just this small thing* which AI seems helpful for is an accident.

Show threadStephen FarrugiaMay 19
I have much more to say but this is so exhausting
Show threadStephen FarrugiaMay 19
these threads always uncover the same what-aboutisms that don't deserve a response
Show threadStephen FarrugiaMay 19
I've yet to meet a "what about?" that hasn't made me roll my eyes
Show threadStephen FarrugiaMay 19
I do have a few "I dare you" ones I'm ready for
Show threadDave AlvaradoMay 19
@fasterandworse saw this today, seems appropriate for this moment.
Show threadStephen FarrugiaMay 19
@dave I thought the middle one was a poor likeness of tim cook at first
Show threadElton Cora (Honorary)May 20
@dave @fasterandworse if they’re not careful, people will start coming around to the joys of being left behind.
Show threadMike @ Misaligned MarketsMay 19

@fasterandworse I think a lot of technology, esp in anglo-capitalist societies, started off harmful and became relatively mundane through regulation and informed use. I say this not out of optimism but pragmatism.

I don't think we should encourage the use LLMs. But in a world where some people are already using them and scoped use cases + alternatives to centralized platforms are emerging, I want to steer as much as possible toward “less bad” if a ban won't happen.

https://zirk.us/@MidniteMikeWrites/116104686552109831

Show threadMike @ Misaligned MarketsMay 19

RE: https://zirk.us/@MidniteMikeWrites/116523605682492305

@fasterandworse I recognize the point you're making. LLM chatbots rely on affective design for user feedback. They inherently have no scope or affordance that makes them “for” anything.

People, however, are building workflows in scoped domains where users don't externalize their cognition and model outputs can be evaluated by some standard.

In so far as these conditions are met I think future harm can be mitigated. But I understand this doesn't change the ethical/sourcing issue, though.

Show threadMeznorMay 20

RE: https://hci.social/@fasterandworse/116602509715514659

...and it is no surprise (or should not be) that the traditionally revenue generating things - surveillance, war, clickbait-y disinformation - are emerging as revenue generating use cases above and beyond any others. The system (and small group of proveably bad people) that created genAI/LLMs/MLs are repeating their dominant patterns of exploitation and harm.

Show threadTxo!?May 19
@fasterandworse to be fair, they are designed for a purpose, that is deceive the user and suck from they as much as it can
Show threadLeendaalMay 19
@fasterandworse i disagree they are designed to establish fashism.
Show threaddcreMay 20
@fasterandworse And LLMs are hardly the best (or even good) at anything. All they produce is mediocre dice-roll text that you take more time editing than doing it yourself entirely..
Show threadKorma ChameleonMay 19
@fasterandworse just burning tokens as required by my manager and her manager, all the way up the company
Show threadKargasMay 19
@KormaChameleon @fasterandworse Yeah, in that situation too.
Show threadSzescstopniMay 19
@fasterandworse The old LLMs, not so blood- and energy-thirsty, were pretty good at translating natural languages. And some expert systems are good at what they were designed for. But they were not really a part of the AI industry we have right now. I use matches for lighting our wood stove. Not for starting a fire in the forest I live in.
Show threadМя ��May 19

@Szescstopni yeah, ML models designed and trained for specific task are pretty useful, and usually are much, MUCH lighter in terms of resources

Using cloud based text generator as a magic tool for everything, however, is not

@fasterandworse

Show threadSzescstopniMay 19
@mo Of course. What is now called "AI" should burn. @fasterandworse
Show threadKlingon SpongeMay 19
@fasterandworse All the small things add up to a big thing.
Show threadFelix 🐊May 19

@fasterandworse i'll accept any and all empirical evidence in favor or against LLMs, but it just so happens the evidence against them is overwhelmingly convincing.

People *have found uses for them*. It just so happens their usage is usually unethical, given that most modern LLMs are trained on stolen data to begin with. As much as the tool is incredibly inefficient at most things, the things it can do we have had trouble designing tools for previously, and with the current hype surrounding them it's easy for people to blow their usefulness out of proportion without considering the caveats.

Bias in their output is inherent in their construction, from the data they're trained on to the people who program their interfaces to filter the results. LLMs clever enough to trick gullible business owners into "replacing" people have now caused real people to have to brave questionable welfare systems until they can find a new job or change industries. AI used to make professionals' jobs easier ends up deskilling the very professional it was designed to help. Generated images become a hollow replacement for human connection and honing a craft, an opportunity to feel efficacious.

The AI problem isn't inherently a problem about their existence or usage; after all, it's just a dumb tool. But it is largely a problem about the context within which they exist and are being used, which ends up being outrageously inappropriate. In our current world, for each solution, it creates a problem

Show threadMatt GoddenMay 20
@fasterandworse "I only wear shoes produced by enslaved child labour when mountain climbing. I wouldn't wear child slave labour shoes all the time."
Show threadAssimilateborgMay 20
@fasterandworse only way is to self-host your own AI thing on your own gaming PC. So either you game or you AI. And quickly learn that AI is way less satisfying when it gets in the way of your gaming.