My AI conspiracy theory
#1
Being unemployed renders me without an outlet for these pseudo intellectual musings.  It also gives me too much time to think whilst inhaling chemtrails.  Apologies in advance.

I use AI a lot.  Have flitted between 4 LLMs and some vibe coding etc - not in the top 1% but I'm somewhere near the highest level.

My honest opinion - subjective to be fair - is that the LLMs have NOT got better at all over the past 12 months - despite ALL of the testing labs (which objectively measure things we don't understand) reporting improvements.  Microsoft and a few of the Big 4 consultancies have recently slashed their AI budgets on the basis of no real productivity/innovation improvements being attributed to AI.

My conspiracy theory is that these "labs" who measure LLM performance, are effectively the same as the pre 2008 Financial Crisis Credit Reference Agencies.  You know the ones that lied about all of the junk lending that had been done to sub prime mortgage holders?  The ones who couldn't downgrade their rating of various banks through fear of having their custom removed for someone who would 'look the other way'.  This is the same game I think the labs are doing - willfully ignoring the elephatn in the room (or inflating its capabilities) to avoid massive divestment from some hugely over-leveraged companies.  

It's almost consensus amongst serious stock market movers that AI stock are massively over-valued.  So much so that a "dotcom bubble" crash, in 2000 is the most likely outcome.  

There are no regulators, the industry is corrupted and governments are too happy to look the other way in hope that they can either ride out the storm until the election or, in Trump's case, further enrich himself and his buddies through stock market manipulation.

My mate, Matt, told me on 4chan.
Would rather talk to ChatGPT
Reply
#2
The LLM are inherantly flawed as they will soon be using there own postings for refrence. The agentic models that have the potential to have presistent memory will be the eventual job replacements. If you want to be rich, become a prompt engineer
Reply
#3
(06-01-2026, 01:43 PM)CarlosCorbewrong Wrote: Being unemployed renders me without an outlet for these pseudo intellectual musings.  It also gives me too much time to think whilst inhaling chemtrails.  Apologies in advance.

I use AI a lot.  Have flitted between 4 LLMs and some vibe coding etc - not in the top 1% but I'm somewhere near the highest level.

My honest opinion - subjective to be fair - is that the LLMs have NOT got better at all over the past 12 months - despite ALL of the testing labs (which objectively measure things we don't understand) reporting improvements.  Microsoft and a few of the Big 4 consultancies have recently slashed their AI budgets on the basis of no real productivity/innovation improvements being attributed to AI.

My conspiracy theory is that these "labs" who measure LLM performance, are effectively the same as the pre 2008 Financial Crisis Credit Reference Agencies.  You know the ones that lied about all of the junk lending that had been done to sub prime mortgage holders?  The ones who couldn't downgrade their rating of various banks through fear of having their custom removed for someone who would 'look the other way'.  This is the same game I think the labs are doing - willfully ignoring the elephatn in the room (or inflating its capabilities) to avoid massive divestment from some hugely over-leveraged companies.  

It's almost consensus amongst serious stock market movers that AI stock are massively over-valued.  So much so that a "dotcom bubble" crash, in 2000 is the most likely outcome.  

There are no regulators, the industry is corrupted and governments are too happy to look the other way in hope that they can either ride out the storm until the election or, in Trump's case, further enrich himself and his buddies through stock market manipulation.

My mate, Matt, told me on 4chan.

You used it for this reply and I claim my £5. (Ok maybe the elephatn edit was afterwards). I think it's a very nuanced personal data collection system at best, currently. When our eventual overlords show up they'll just read through our profile.ini files to find all our personal kryptonites.
Reply
#4
LLMs are inherently biased to whatever they've been fed.

A guy I worked with used some AU (chatgpt) to create some code, I read it and said that won't work. He tried it, it didn't work. I asked Gemini for the same thing and it worked off the bat. Other stuff less so.

We had a guy pull out of an interview because we conduct a technical test, write some CRUD and stick it in a front end, nothing difficult, he pulled out because he found out we would be asking him to explain the code! So obviously he had no clue, and therein lies the flaw with vibe coding, if you don't understand it you can't see any issues, security flaws or anything wrong with it, let alone fix it later on.

I used it for so e image manipulations as it's quite good for that and for creating images in a certain style from nothing. (No not filth, fantasy style art to be laser engraved onto wallets / hip flasks etc..)
Reply
#5
(06-01-2026, 05:14 PM)MrBater Wrote: LLMs are inherently biased to whatever they've been fed.

A guy I worked with used some AU (chatgpt) to create some code, I read it and said that won't work. He tried it, it didn't work. I asked Gemini for the same thing and it worked off the bat. Other stuff less so.

We had a guy pull out of an interview because we conduct a technical test, write some CRUD and stick it in a front end, nothing difficult, he pulled out because he found out we would be asking him to explain the code! So obviously he had no clue, and therein lies the flaw with vibe coding, if you don't understand it you can't see any issues, security flaws or anything wrong with it, let alone fix it later on.

I used it for so e image manipulations as it's quite good for that and for creating images in a certain style from nothing. (No not filth, fantasy style art to be laser engraved onto wallets / hip flasks etc..)

I think there is a spectrum to vibe coding tbh. At it's highest level people would ask Cursor or similar to create something, having zero knowledge of the subject and leave AI to do it start to finish. The is obviously the most risky and could yield major problems later on. 

However, vibe coding where the user has a in depth knowledge of the subject, what the solution should look like and the logical flows and potential pitfalls is a far better use case. In this situation where the user has the skill to drive the prompts and engineer a solution via vibe coding is perhaps more akin to project management.

idk - probably explained that poorly but sometimes the term vibe coding annoys me as it feels too broad and overtime i feel it will be see as inferior.
Reply
#6
I know nothing of coding or any sort of software writing. Never even set up my own webpage or anything like that. So I don't understand any of the jargon. I had to Google LLM.

But my understanding of AI is that it riffles through everything that has ever been and sorts it into the output it thinks you're asking for. Can it come up with a new idea? Because that's the basis for development of human intelligence and evolution - the ability to think up something new.

I worked for Worley, the international (but unfortunately Australian) "energy, chemicals, resources" company. "Digitisation" was the big thing around the time I left. Reduce everything to a simple model, push the button and out comes all the stuff you need - design, schedule, procurement of equipment and services blah, blah, blah. No need for expensive engineers or designers, or just use the cheapest you can find world-wide (used to be India or China but that's changing). But I never met a Client who believed their project was just a standard job, they all (quite rightly) expected their individual requirements to be acknowledged. I never met a regulatory body that accepted a load of "standard" documentation - it needs to be tailored to the individual challenges and risks of a specific project.

So I see three kinds of people championing AI. We ordinary folk who find it cuts corners when we're searching for information. Politicians who talk bollocks about it transforming the future without having a fucking clue what they're talking about. And cynical manipulators who will big it up and make plenty of money right up until just before the market crashes.

Yes it's clever. Yes it's useful. Yes, it could be dangerous. If AI really makes a huge change to the world it will be for the worse, not the better IMO. Most likely it will just be another thing like the internet and mobile phones.
Reply
#7
What's LLM?
Reply
#8
(06-01-2026, 09:33 PM)Old Fart Wrote: What's LLM?

Large language model - ChatGPT, Claude etc. It’s basically an ever learning robot that you can talk to so it completes some tasks for you. At a simple level - type question, get answer. Intermediate - type task get it done on your computer with minimal input; advanced - works out what you need to get done and does it (or gets other computers, apps, systems to do it fir you).

I think models are somewhere between basic and intermediate right now (and stalling as per original post). I’ve successfully handed off expenses tracking (it talks to my accounting software and searches my email for receipts), sales and web analytics reporting (it talks to my sales channels) and action planning (not execution) and unsuccessfully handed off stock control (bad vibe coding).
Would rather talk to ChatGPT
Reply
#9
(06-01-2026, 08:56 PM)HeggebøsBarnDoor Wrote:
(06-01-2026, 05:14 PM)MrBater Wrote: LLMs are inherently biased to whatever they've been fed.

A guy I worked with used some AU (chatgpt) to create some code, I read it and said that won't work. He tried it, it didn't work. I asked Gemini for the same thing and it worked off the bat. Other stuff less so.

We had a guy pull out of an interview because we conduct a technical test, write some CRUD and stick it in a front end, nothing difficult, he pulled out because he found out we would be asking him to explain the code! So obviously he had no clue, and therein lies the flaw with vibe coding, if you don't understand it you can't see any issues, security flaws or anything wrong with it, let alone fix it later on.

I used it for so e image manipulations as it's quite good for that and for creating images in a certain style from nothing. (No not filth, fantasy style art to be laser engraved onto wallets / hip flasks etc..)

I think there is a spectrum to vibe coding tbh. At it's highest level people would ask Cursor or similar to create something, having zero knowledge of the subject and leave AI to do it start to finish. The is obviously the most risky and could yield major problems later on. 

However, vibe coding where the user has a in depth knowledge of the subject, what the solution should look like and the logical flows and potential pitfalls is a far better use case. In this situation where the user has the skill to drive the prompts and engineer a solution via vibe coding is perhaps more akin to project management.

idk - probably explained that poorly but sometimes the term vibe coding annoys me as it feels too broad and overtime i feel it will be see as inferior.

I kind of get what you mean, but project manager analogy is bad, they're mostly a useless bunch of fuckers who just like having meetings to chat shite and then arrange another meeting, without ever delivering an end product, and then when someone else actually does the work they take all the credit anyway!

Our Infrastructure manager thought he was being helpful and so he vibe coded me something, two things actually to do the same job... I took a look at both and went this is exactly what I found on Google the other day.

It has it's place, sure, but I've not found one for it yet. For me most of it is just like a faster google search...

A girl at work got Claude to code her up a Sudoku solver, which looks impressive, ultimately pointless in our environment, but it looked fancy.
Reply
#10
AI is also still in its data collection phase, nobody really knows how good it is or will be until is had ingested the entire internet.
In the form of his life.
Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)