Artificial intelligence still has some way to go

if you’re wondering what the product looks like: https://learn.microsoft.com/en-us/azure/ai-services/openai/use-your-data-quickstart

My professional world is largely technical writing, in a competitive context.

An AI that goes out to the internet and does writing-by-google is worse than useless, because every competing company can submit exactly the same content.

The tools that I think have more potential are those that do NOT go outside, and just parse my existing library. But... I could just do that myself.

I think the risk isn’t that human writers go away, it’s that all writing becomes this consumed-by-ai technical writing specifically structured in ways that make sure the LLM doesn’t fuck it up when it regurgitates it

chatgpt gets extra functionality quite often - now it does pictures, now it reads PDFs, now a bit of data viz, now it searches the web. And sometimes a bit of it goes, and then it comes back… it's a confusing platform tbh

The MS thing sounds like it overlaps with GPTs?

yes. they’re a close licensee of chatgpt, kind of a second-party provider, and possibly on the way to being a licensee that actually does most of the development as part of a consortium

I think the cynical take I've landed on is that the product of AI will kinda remain half-assed garbage but that people will slowly adjust to the garbage until it becomes an acceptable alternative to human-generated product in many contexts. Like I can see the writing staff of every show on CBS being replaced by AI without anyone really noticing or caring.

fwiw human language is only one of the things LLMs are decent at. they’re basically pattern recognition and inference tools

https://arstechnica.com/science/2023/03/large-language-models-also-work-for-protein-structures/

afaict one of the teams I support is using a sieve-like approach to run multiple LLMs and other ML models to infer protein structures in the way mentioned in this article

xp maybe! I think the crass commercialization aspect is that we’re going to get more sophisticated versions of chum. those nonsensical chumbox articles at the bottom of pages that Clickhole satirizes will start looking different. episodic tv of the NCIS or w/e variety is basically background noise chum already and it’s easy to imagine that being generated

the issue is that there are no new ideas and you still have to have something to pull from, whether it’s a decade of scripts or just throwing all of the past tv shows in a blender plus something new. to an extent all human endeavors are a mishmash of existing material but it’s disarming to think that the AI could eventually build new stories off of the AI’s previous output

i dont think this thing will be able to write a tv show but maybe it will be able to write some of the stuff no one cares about

https://i.imgur.com/LTZ9dGM.png

yeah I think grey goo content world is already halfway here - I see more and more articles that (I think) are clearly chatgpt generated. There's just no cap on feeding it crap questions or a text to summarise and dumping the output straight to the web.

I mean this is zero effort, 5 seconds of typing and 30 secs of waiting for output. It's perfectly sensible and boring as shit.

User
why does a messageboard like ilx survive in this day and age
ChatGPT
The survival and continued relevance of a message board like ILX (I Love Music) in the current digital era can be attributed to several factors:

Niche Communities: ILX caters to a specific interest group - music enthusiasts. This niche focus creates a dedicated community where members share a common passion. Such specialized communities often survive and even thrive despite the prevalence of larger social media platforms.

Quality of Discussion: Forums like ILX often offer a depth of conversation that is harder to find on mainstream social media platforms. The format encourages longer, more thoughtful posts and discussions, which can be appealing to those seeking more than just quick, superficial interactions.

Sense of Belonging and Identity: Long-standing members of ILX have likely developed a strong sense of community and identity within the forum. This can create a loyal user base that continues to participate and contribute over time.

Simplicity and Familiarity: For some users, the straightforward, text-based format of traditional message boards is preferable to the more complex interfaces and algorithms of modern social media. Additionally, long-time users might prefer the familiarity of the platform and its community norms.

Archival Value: Forums like ILX often have a rich archive of past discussions, which can be a valuable resource for information and historical perspectives on various topics, particularly in niche areas like specific music genres or artists.

Avoidance of Mainstream Social Media Pitfalls: Some users might prefer message boards to avoid the downsides of mainstream social media, such as data privacy concerns, overwhelming amounts of content, or negative mental health impacts.

Moderation and Governance: Message boards often have a different approach to moderation compared to larger platforms. This can sometimes create a more controlled environment where discussions are more respectful and productive.

In summary, the survival of a messageboard like ILX is likely due to its ability to offer a unique, niche community experience, fostering in-depth discussions and a strong sense of belonging among its members, away from the pitfalls of mainstream social media.

on the web theres so much seo driven stuff that might as well be ai already, its prob where ai picked up its terrible style even

yeah, the grey goo idea does grab me because of playing with the damn thing but I don't know if it's actually just the shit tide rising a little higher, or full tsunami because barrier to entry is now unbelievably low. + once the compute power gets properly priced in and venture money thins, maybe it will still be cheaper to offshore SEO content trash? Although it looks like a tuned 70B llama 2 (ie sort of open source) could handle the shit-production job at scale for peanuts.

people are already pulling that shit. pretty sure I mentioned it before, but searching for specific topics that seo ad farms have picked up on gives you garbage results. video game walkthroughs were the one I was encountering, but you occasionally get career-specific info too

it’s a dead giveaway when they fuck up and reuse the images, but half the time, it’s just a mostly-coherent webpage that’s a mechanically reworded version of someone else’s content but with double the ads and no attribution

I can't wait until every article is written in the format of a 9th grade paper about the Civil War. Gimme proper introductions and conclusions or gtfo, I says.

I was finding guides to using the chatgpt api that were written by chatgpt.

I twigged when I just decided to ask chatgpt

It's like Human Centipede but where ChatGPT's mouth has been sewn to its own anus.

I would ask chatgpt for a fuller description of the chatgpt centipede, possibly with a dall-E illustration, but unfortunately it is down

The robots have not thanked me, as far as I know. And when I am unemployed, probably next month, they will not pay my mortgage. So forgive me if I am little bit bitter about that.

― Oh I believe in Yetis' Day (Ye Mad Puffin)

ai is going to wind up taking more and more jobs... not just directly! the ones that are left require an ever-changing set of skills... it's really hard for me to continue to keep up with what's considered "employable" these days. i am seeing a lot of people in the same boat. people get laid off and finding a new job is just really difficult these days. a lot fewer jobs. a lot less money. plus, a lot of my friends are "disabled" and not able to work. the economic and social values i was taught was that people who aren't actively working are a drain on society, a burden. if they're suffering, well, it's their own fault. that's what i was taught. those are the values i see being put into practice.

these aren't values i particularly agree with. i don't think that these values are... a viable basis for a society, on a long-term basis. i don't have the power to change those values. it's a major source of ongoing stress in my life. i employ a lot of radical acceptance. i'm getting a lot of opportunity, these days, to practice radical values.

anyway i'm not going to throw one of my shoes at gpt. those things are fucking expensive, i need me some shoes.

American capitalism is based on a covert “deal.” Work hard and you will find a niche where you “provide value to shareholders” and then you’ll get paid so you don’t starve. This deal was never actually true, but for many people it looked more true in the past than it does now. And so the question is what will happen when more people acknowledge that the old covenant is *really* broken this time. There is no path to living in comfort and dignity for many.

So much of politics and ethics in america is built on what you described kate. The idea that if people were willing to work they would at least prosper moderately. If chat gpt destroys, say, 80% of lawyers’ jobs all of that kind of thinking is going to seem laughable, even to the formerly comfortable middle and lower bourgeoisie.

It could be a revolutionary moment. More likely America will just become colder and angrier.

im gonna wait til ai demonstrates 5% of the competency required to do a job before i start worrying about it taking them

Best case scenario, people stop valuing human beings for what they can “provide” and start appreciating them for their intrinsic value. The kantian imperative. And then we build new economic and political structures around this humanistic framework. The odds seem long for this one though.

also worth noting that software and indeed technology generally has been "taking jobs" for a long time, are there fewer jobs now, not really, are the current jobs worse, yes in a way, specifically in the way that they pay less offer less security etc, is that because the jobs technology took were inherently better, was file clerk a somehow great job, no, the issue with downward mobility amongst working people is of course one of worker power, and corporate power

if you want the kantian imperative you have to join a union, then you can ask them for it lol

interesting question is ask people what job their grandfaters did if they owned a house if their grandmothers worked, tho this may be more of a great grandfather question for some people at this point, couple memorable answers ive gotten for home owning sole bread winners were worked at a news stand and donut maker

I am in a union xp

nice

My paternal grandparents owned a house in rural Indiana, where they ran a small restaurant. My maternal grandparents lived across the street from them. My mom's dad was the town doctor.

I visited recently because my uncle passed away. The lawyer who was handling the estate said that my grandfather had delivered her as a baby.

my grandfather was born on the farm our ancestor bought with money from the gold rush, america gotta love it baby

More likely America will just become colder

Wait I thought the leading doom scenario was about things getting warmer.

ayyyyyy

My grandfather made suits, and there's a family story that he won a contract to make uniforms for the Chicago Police in a poker game.

AI only does what you ask it... it's not quietly thinking of new products, new films, new ideas; it just responds to input and barfs out a facsimile

There is no force that drives the fuse

I agree. The question is how many jobs require simply regurgitation of that kind. I personally don’t know.

I don't disagree in principle - certainly that human intervention is currently needed, a lot - but I'm a bit more pessimistic about how much of the world of work straight statistical mimesis can eat, especially as gen ai improves.

I agree. The question is how many jobs require simply regurgitation of that kind. I personally don’t know.

the stumbling block for the automators has been that people

It’s a paradox though because while automation causes individual firms to save money and increase profits, in the end it causes the economy to shrink.

I think. Wouldn’t it cause a crisis if unemployment drastically rose? Who would be spending money on goods and services?

https://en.wikipedia.org/wiki/Tendency_of_the_rate_of_profit_to_fall

i mean the economy has grown quite a bit since automation started

My understanding of this is incomplete (intentionally, for reasons already discussed). But didn't Open AI stop collecting new information a few years ago?

As a result, AI's assumptions are incomplete at best, and at worst will just get more and more outdated.

To be useful in future, it will need more humans to fuel it and give it more IV injections of that sweet, sweet, content. Yeah baby (I hear it saying), feed me content.

We are all Seymour from "Little Shop of Horrors."

what if the ai apocalypse is just the machines forcing us to make more content, which we were doing anyway

Trenchant

I will go in to work tomorrow and cheerfully type for my robot overlords. Because it's not like I have a lot of other choices.

Personally I am too old to shift gears and become, like, a useful member of society or whatev

If it's dumped onto the internet anywhere it's all fodder for LLMs, because copyright doesn't apply to how AI munches on its manifold inputs, digests them, and creates its mindless homogenized sentences. We ilxors converse openly about whatever interests us and consequently we are ingested. It even applies retroactively and involuntarily to conversations long before these chatbots existed.

My understanding of this is incomplete (intentionally, for reasons already discussed). But didn't Open AI stop collecting new information a few years ago?

The version anyone can use has a training cut-off date (currently April 2023). But behind the scenes newer models are being retrained & they'll be up to date, tho I don't know what the deal is with the presumably significant problem of it being fed its own shit.

Can users customize an ai dataset? I’m curious about an AI trained solely on the ilx archives. Would they be chill? Would they be an asshole? Would their prose still seem like an alien wrote it while wearing a human suit?

AI Centipede

xp

didn’t I post a few times about how people add data sets and knowledge bases to ChatGPT/llama/etc or am I taking crazy pills here

fwiw the entire shtick we’re all doing where we ask it questions and assume it’s giving facts of some sort is just a side effect of how it works. it’s overall irrelevant, or supposed to be, to their business model that ChatGPT spits out things resembling facts, afaict. it’s just a chatbot that can string together language in a human-like manner. GPT as a product that you license and plug your own factual data in is supposed to be the value add