Post by @theodoraward@wandering.shop

RE: https://tech.lgbt/@lettosprey/117534851823121382

one of my big missing-stuff-but-i-find-it-useful Theories i am kicking around these days is that silicon valley realized -- first with windows, then with the iphone -- that if you wanted to really generate money, don't just focus on software: control the entire computing environment. with AI they set out to make an entirely new, thoroughly proprietary computing environment, but they didn't quite realize that the environment itself isn't where the profits come from

Lett Osprey :v_pan: :v_enby: 🍉 :therian:

@lettosprey@tech.lgbt 8 hr. ago

My non-existent AI usage is quite rooted in the fact that I don't want to support this resource draining theft monster, but if we put that aside, if that wasn't an issue, then what?

I hear about what people use AI for, but I cannot figure out how any of it would be helpful for me.

Sure, I could ask it to write some code for me, but I would spend more time reviewing the code and making sure it did exactly what I wanted, than just writing it myself.

Write a grocery list? I hear people use it for that! .... I WANNA DECIDE WHAT TO EAT MYSELF THANKYOU!

I could use it to help me write music! ... that... I would not have written then, it would not be mine, I would not feel like I had accomplished anything, I cannot lie to myself and think "I made this!"... I either make it myself or I can just listen to music other people made.

Home automation! ... What??? I wanna know that what I want to happen, happens when I want it to happen!

Call me old fashioned, but I am at a loss for what AI could help me with.

2 Boosts

0 Likes

Theodora Ward

@theodoraward@wandering.shop 3 hr. ago

This list is -- well, it's not how i'd go about doing this research -- but it speaks to this, I think. The two killer apps for LLMs have been chatbots and code generation. Most suggested use cases for chatbots are either weird little things around the edges; besides that, it was basically inevitable, I think, that sycophantic chatbot usage would drift ELIZAward (in a way that, to me, represents a public health hazard)

Massive list of how people are using gen AI from Harvard Business Review(source: Filtered.com) . The caption: "Have people found ways for generative AI to help liughten their workloads, increase their productivity, or think through problems in new ways? To understand how individuals are using the technology, researchers mined web forums like Quora and Reddit, filtering through tens of thousands of posts to identify 100 different use cases, which they then organized into six themes: personal and professional support, content creation and editing, learning and education, technical assistance and troubleshooting, creativity and recreation, research, analysis and decisionmaking."

I can't transcribe the whole thing but the top ten are: therapy/companionship, organizing my life, finding purpose, enhanced learning, generating code (for pros), generating ideas, fun and nonsense, improving code (for pros), creativity, healthier living — Massive list of how people are using gen AI from Harvard Business Review(source: Filtered.com) . The caption: "Have people found ways for generative AI to help liughten their workloads, increase their productivity, or think through problems in new ways? To understand how individuals are using the technology, researchers mined web forums like Quora and Reddit, filtering through tens of thousands of posts to identify 100 different use cases, which they then organized into six themes: personal and professional support, content creation and editing, learning and education, technical assistance and troubleshooting, creativity and recreation, research, analysis and decisionmaking." I can't transcribe the whole thing but the top ten are: therapy/companionship, organizing my life, finding purpose, enhanced learning, generating code (for pros), generating ideas, fun and nonsense, improving code (for pros), creativity, healthier living

Theodora Ward

@theodoraward@wandering.shop 3 hr. ago

LLM text gen is technically extremely impressive. If you presented it tome just, like, as a weird cool thing, my mind would be blown. But I think c-suites underestimated the extent to which the limitations of the tech -- most fundamentally, (a) the costs and (b) the disproportionate impact on "workflows" and use-cases of probablistic errors like hallucinations. They heard that search fucks up 4% of the time and thought hey yeah that's close enough

David Gerard

@davidgerard@circumstances.run 3 hr. ago

@theodoraward 100%. it's amazing research, not machinery that can be relied on.

Theodora Ward

@theodoraward@wandering.shop 1 hr. ago

@davidgerard i remember the message board thread where we all played with gpt-2 was so funny i was literally crying. then the humor faded, because the surprise wears off fast, because it is a technology built on recapitulating patterns, and we all moved on. i know pretraining was still scaling up so the shock would've renewed for at least the next gen or two, but i still think about that experience as a basically proportionate response to improvements in the plausibility of text generation

1 more reply

Theodora Ward

@theodoraward@wandering.shop 3 hr. ago

But probablistic error is a qualitative problem, at least in higher-stakes environments like workplaces. (If not because the work matters, then because its quality bears, at least theoretically, on your livelihood.) Sometimes a guess is probably fine; but when I'm actually researching something, I unconsciously rely on a kind of chain of citations -- find a website that seems trustworthy, assess it again when I see what it looks like, check their sources as needed -- that LLMs obfuscate

Theodora Ward

@theodoraward@wandering.shop 3 hr. ago

They also underestimated the cultural backlash and the medium- to long-term cognitive effects some people, at least, experience from engaging with text generation, among other things. They thought -- probably still think -- they were inventing an entirely new means of consumer computing; they thought the applications would come. That's why they're frantically trying to backfill with goofy startups that just subsidize text-gen API pulls

1 more reply