OP
OP
briantrumpet
Timewaster
Cookie Monster has opinions.
AI fails
- Thread starter briantrumpet
- Start date
Page may contain affiliate links. Please see terms for details.
AndyRM
Elder Goth
He isn't wrong.
stowie
Well-Known Member
I love the bit "Me horrified. Me have unique voice. And now me not sound like me. Me sounds like Matt on Linkedin".
Linkedin is the "ground zero" of AI enshittification. It has turned every feed into a beige like wall of platitudes. I mean it wasn't great before but at least some posts sounded a bit different from others and some posts were interesting. Those posts are still there but now buried under a tsunami of crapola. I suspect those running social media platforms initially loved AI as it allowed instant "content" posting. More content, more eyeballs scrolling more content for longer, more ad revenue. But I think it is backfiring as people get bored with it. I for one wish for a return of the over the top posts about overcoming "extreme adversity"* to hit some arbitrary business target. At least they were amusing.
*"Extreme adversity" normally being some very minor first world problem as opposed to actual extreme adversity like lack of access to potable water or civil war.
Last edited:
stowie
Well-Known Member
Back in the early 90's as a junior engineer I had a very minor role in a project investigating neural networks for image and sound classification. The models used would not be totally unfamiliar to someone using models these days for the same task. The issue was not a maths or science one. It was engineering. The compute power needed even for inference was so big in volume, weight and cost that practical applications were difficult to realise.
30-odd years later the increase in compute power has helped that engineering problem hugely.
Now it is a business problem.
We have Coreweave - the datacentre provider for AI compute - finding its bond yields hitting 15% or higher. Anthropic is offering yields of 10% or more to raise funds for compute build out. Not only this, but the big players such as NVIDIA are circular funding with backstops. This type of activity doesn't really scream that investors believe all of this is now low risk.
I use Claude Code every day. It is probably the biggest productivity boost I have witnessed since the internet. The things it can do are frankly amazing in my area of technology - which has been later adopting LLMs in development than software. Only yesterday, I ran three what-if scenarios on a design to test some ideas. Without Claude Code, this each one would have taken days if not a week or more. All three were devised, run and tested within an afternoon. The problem is not usefulness in certain industries. The problem is how much this can scale and how much people are willing to pay for it. In technology I suspect the scaling is relatively modest but willingness to pay is quite high. For general use like my family use ChatGPT I suspect scaling is very big but willingness to pay is zero or close to zero. Going back to google search isn't a huge backwards step, just a bit annoying.
The AI failure isn't really the technology itself but the value placed upon it and the push to make it all pervasive in our lives. Back in the 90's it was obvious that the cost and size of infrastructure needed massively outweighed the value. Both sides of the equation tipped closer to each other in recent years but still with a difference that was filled by optimism, boosterism and accounting and finance that is - lets say imaginative.
I mean the fundamental question I have is that my little part of the industry finds it incredibly useful and productive. But what is the point of AI generating vast quantities of books, music and films / TV? Surely these are human endeavours based around story telling and communication which are a human construct, created by humans for humans. Where does AI land us? An ocean of material that may or may not be as good as human creation but at such an output and volume that the whole experience is devalued. There is something just a bit weird about the concept.
30-odd years later the increase in compute power has helped that engineering problem hugely.
Now it is a business problem.
We have Coreweave - the datacentre provider for AI compute - finding its bond yields hitting 15% or higher. Anthropic is offering yields of 10% or more to raise funds for compute build out. Not only this, but the big players such as NVIDIA are circular funding with backstops. This type of activity doesn't really scream that investors believe all of this is now low risk.
I use Claude Code every day. It is probably the biggest productivity boost I have witnessed since the internet. The things it can do are frankly amazing in my area of technology - which has been later adopting LLMs in development than software. Only yesterday, I ran three what-if scenarios on a design to test some ideas. Without Claude Code, this each one would have taken days if not a week or more. All three were devised, run and tested within an afternoon. The problem is not usefulness in certain industries. The problem is how much this can scale and how much people are willing to pay for it. In technology I suspect the scaling is relatively modest but willingness to pay is quite high. For general use like my family use ChatGPT I suspect scaling is very big but willingness to pay is zero or close to zero. Going back to google search isn't a huge backwards step, just a bit annoying.
The AI failure isn't really the technology itself but the value placed upon it and the push to make it all pervasive in our lives. Back in the 90's it was obvious that the cost and size of infrastructure needed massively outweighed the value. Both sides of the equation tipped closer to each other in recent years but still with a difference that was filled by optimism, boosterism and accounting and finance that is - lets say imaginative.
I mean the fundamental question I have is that my little part of the industry finds it incredibly useful and productive. But what is the point of AI generating vast quantities of books, music and films / TV? Surely these are human endeavours based around story telling and communication which are a human construct, created by humans for humans. Where does AI land us? An ocean of material that may or may not be as good as human creation but at such an output and volume that the whole experience is devalued. There is something just a bit weird about the concept.
swee'pea99
Regular
Brings to mind a story my dad once told me he'd heard from a Yamaha chap - this was way back in the day - about the Brit who'd been trying to interest motorcycle manufacturers in this 'computer-modelling' thing he'd invented that could analyse gas flows and exploit that data to reduce fine-tuning of cylinders and the like from a time-honoured trial & error months of metal-grinding to a few minutes on a computer, and they looked at him like a kid who was trying to tell them - them! - how to build motorbikes. And said thanks very much and the door's that way. And after battering his head against one wall after another for a year or three he went to see Yamaha in Japan, who said "Come in..."
I suppose one clear difference is that lots of people want, and are prepared to pay for, a better motorbike, but not that many people want to run what-if scenarios on a design to test ideas. And it's hard to see what it can do that many people will be prepared to pay for, or that they won't be able to get somewhere else for free, with so many 'intelligent' agents competing to help for free. If you won't do it then your sister will...
OP
OP
briantrumpet
Timewaster
A good summary, I think. It's very good at certain things (and will get better), is OK at others (if highly inefficient), and deeply destructive in others (not least human cognition, if over/misused). I think the parallels with other things that have been over-hyped and over-used at launch will become more apparent at time goes on, and as the true cost (as the circular financial mechanisms employed run out of road) is realised.
stowie
Well-Known Member
I do have to temper my cynicism by remembering that once upon a time I thought the Internet would be useful but probably a bit niche...
However, the road to tech bro hell is littered with technology "paradigm shifts" that solved a problem no-one had. And outside my work, I consider what I use LLMs for and it is essentially as a super-sized search engine. The value of this? Maybe not as much as AI companies would like (or need to charge). So the value may be in work related activities (in unrelated fields, I see how law and medicine could make real productivity gains with versions of LLMs - basically any work involving a lot of information to process which has verifiable data points). Does this validate the extraordinary valuations of the AI companies or justify NVIDIA being 8% of the S&P500?
OP
OP
briantrumpet
Timewaster
I was calling it "Google with knobs on" from the early days (and sometimes use it like that). But it's worth remembering that just old-skool Googling requires both skill and knowledge to get the best results (working out key vocabulary and how to use that for prioritising search results), so the 'with knobs on' element is risking human users not developing the framework of understanding to then question the results that LLMs confidently churn out (without any understanding at all).
Makes me think a bit of this piece of music
View: https://www.youtube.com/watch?v=3A0JfsJt98Y
The eponymous sorcerer's apprentice was imagined by Goethe:
"The poem begins as an old sorcerer departs his workshop, leaving his apprentice with chores to perform. Tired of fetching water by pail, the apprentice enchants a broom to do the work for him, using magic in which he is not fully trained. The floor is soon awash with water, and the apprentice realizes that he cannot stop the broom because he does not know the magic required to do so."
https://en.wikipedia.org/wiki/The_Sorcerer's_Apprentice
I think that's where we are with LLMs and AI, except we haven't got the experienced old wizard in the story who can just stop the stuff when it gets out of hand, and no-one understands any of the code its written, and not enough understanding of specialist subjects to question the LLMs when they churn out dangerous codswallop.
Pross
Guru
On the subject of AI in LinkedIn, my feed seems to be over-run with posts not linked to any of my connections using AI graphics to ‘mansplain’ various elements of my industry. Assuming the people posting are real rather than bots they really seem to like to take a Janet & John approach but even then they tend to be ridden with errors. It’s always been a pretty rubbish platform but it’s fast becoming unusable.
AuroraSaab
Epic Member
My kids tell me Pinterest is now unusable as so much is AI generated. One result of it is people going into services like tattooists and florists with a Pinterest photo and being told 'It's literally impossible to do that tattoo' and 'Those flowers don't exist'. There's not much in the creative world that AI won't spoil eventually.
OP
OP
briantrumpet
Timewaster
This is a pretty bleak viewpoint on AI in Higher Education from a professor at Boston University.
https://openquestionsblog.substack.com/p/ai-is-killing-higher-ed-heres-how
https://openquestionsblog.substack.com/p/ai-is-killing-higher-ed-heres-how
Beebo
Squire
Pink Panther in France is AI generated.
https://www.bbc.co.uk/news/videos/c6y93ne31gvvo
https://www.bbc.co.uk/news/videos/c6y93ne31gvvo
OP
OP
briantrumpet
Timewaster
'Too good to be true' (and no photographer credited) should have been a red flag, I guess. But I fell for it. I didn't see a video, just a still, but will go back and see if there are any tell tell signs, such as the panther having too many fingers...
stowie
Well-Known Member
A possibly even bleaker view. An intern working at my company doing an engineering course says that the lecturers know LLMs will be used so pile on the homework to the point that not using LLMs is not an option - there is simply too much. It seems the wrong answer to the wrong question.
