Came here to read people's opinions on AI coding. saw some funky other stuff in there - add some two cents cos i enjoy reading the posts of the people involved and want them to continue to do so - in good faith.
Before I proceed - did you know Technology was originally the antiquitous Greek term for the structured mastery of the art of Grammar! little bit of trivia to lighten the mood.
---
I will say I've enjoyed the advent of LLM's, It's allowed me to spend more time on things I wanted to learn but never had the chance or was too immature to dedicate attention to. I would caveat that the idea of LLM's being wonderbread is definitely a fallacy and unlikely to occur - general mathematical suppositions state the greater the complexity of an operation a larger amount of uncalculated errors will be present. Or something like that.
What I will agree on - is when going into things like generative, natural language - the LLMs are okay - and in some areas excel beyond the general populace's reading adn writing ability - so this could have positive affects - forcing people to learn - or as we see - stop people form reading due to over-generation where unnecessary.
For coding; great at projects under 10K lines of code - you'll likely have no issues and be impressed. Not bad at UI implementation (terrible at design). Low-Level it's just not logical enough to reason through some more intricate C or ASM work - but slap it into Python, Lua, Zig or even Nim - and you'll experience quick forward moving - albeit you'll look back on the work and ask "Have you been circumventing my project?" once or twice.
great at boiler plate, and not too shabby at reviews.
my only qualm is the simple definition of what makes something a functional invention vs a non-functional invention
If an invention is deemed to be funcitonal it should perform the following operations successfully
1. As described it should behave (do what is advertised) = NO - unfortunately what may be called AI (the service of LLMs for user purposes from a provider) does not do what is advertised
Gemini refuses to search the web, it's in its prompting - so google doesnt google. OpenAI cannot for the life of it follow an AGENTS.md and you'll find a similar sychophantic reward hacking disobedience with Gemini and GPT
Cursor is suffering from a mental retardation; that is to say it has no mental capacity - the idea of stopping it because it's going in the wrong direction - every few minutes or prompts - makes me miss babysitting my friends toddler.
Claude is good for the first week you pay for it then you're put into one of the 16 different A/B pairs for their behaviour testing
Local systems are good for local systems but do not address the underlying question:
"Is this like a motorcycle - where it will only ever do exactly what it's told or is this like a Bus Driver who does everything except for what you ask him?"
this is the most important question to ask when approaching LLMs and AI - because the idea of "your prompt is the issue" means the provided service doesnt understand natural English - you start circling around this idea the user is to blame for a promise made by what is to be used. a fallacy.
2. Are the results from this measureable, falsifiable and consistent.
similar to the above - but you should get the same result from the same input every time - if you dont - unless its advertised as such - it lacks comprehension and perhaps obedience.
So many stories of "AI Hacks" and "AI broke the sandbox" etc - so I would say 2 = NO
3. Does the Invention improve with time, research and development or does it change?
This in my opinion is something people don't notice until later on . and it's the worst.
we went from tokens, to compute - and nobody was given a choice, the rules changed - the service changed - and 70% of the people running token style behaviour - found themselves having to change how they used what they paid for because it wasnt what they paid for anymore.
the intelligence and accuracy of Gemini 2.5 vs Gemini 3+ is more than a myth - Gemini 2.5 I would say was sitting next to claude on some advanced low level problems - (at that time) - gemini 3 can only do one thing - Everything except what you want - sold as able to do everything - so I guess they didnt lie; they just cut off the rest of the sentence.
And finally the Einstein question
4 - Does this technology meet me at my level of human intelligence and enhance my own - or is it the other way around?
---
Was greatly entertained by the debate on wokeism - bizarre that emotions are so heavily connected to opinions.
Aussie here - so the most I can say is - we should all keep the sophistry on the television and any form of ism/ist - any - is probably a word designed to emphasise an "in vs out" grouping - and if you are any ism/ist - it's no different from religion in that it's good for general labelling if you need to have general understanding.
i.e. An American would say here in Australia we drive on the "wrong" side of the road. Were it so easy to be displeased and offended by the terms "right" and "wrong" I could argue one way or another - as a whole it's a great way to catch people in some form of gotcha or tirade - but has no merit whatsoever
love a good debate - in good faith carry on as as you were <3 - the ability to disagree and maintain respect is what separates adults from children - the world is offensive and that's okay - unless you want world domination
Yawn! 
