Having lived damn near 4 decades at this point been extremely lucky to watch the steady evolution and integration of all sorts of technology into our lives.. From the old commodore 64 days as a 4 -5 year old to the dial up days in my teens to seeing smartphones completely modify how society operates then watching automation and AI after that... It's been a fucking wild viewpoint for sure.
Back when I started programming in my mid teens shit like chat bots were certainly a thing, but they weren't nearly as knowledgeable or capable of the shit our LLMs are today.. It's quite exciting seeing people get into software engineering with the aide of AI, I truly am happy ot see more people diving into the field.. However the amount of apps I see getting launched daily made byt people who frankly have never engineered a system in their life with the help of LLM coding is starting to make me realize we may be seriously overestimating the capabilities of these coding models.
In software engineering there are 1000 ways to do some functions or methods.. But it may in fact be inefficient or even a security risk to do it like... 980 of those 1000 ways. just because "it works" does not mean it should be instantly offered as a service to people. there is an incredible vacuum of quality assurance and due diligence bug testing on these mostly AI written applications.
At absolute BEST our current publicly available code LLMs are about as skilled as a2 - 3 junior developer with extreme confidence and unmedicated ADHD. No, the LLM doesn't understand the scope or security requirements of your project.. Yes it can be used to generate code that runs, but just because it runs does not mean it's good to go.. People are skipping entire fucking phases of project/product development and then having meltdowns online after someone pwns them for using outdated or malicious libraries or having dependencies such as old SSL versions that leave systems wide open to cyber attacks.
AI coding assistance is a great way for people to get into programming, do not get me wrong.. But unless you truly understand the fundamentals of what you are doing and what must be done in order to achieve your functions securely, you really shouldn't be relying on something that ultimately is autocomplete all jacked up on linear algebra..
Decent Coding Models: Codestral, codegeex, deepseek-coder-v2, claude-sonnet3.5
Smaller lower quantization formatted models can be useful for writing small annoying simple functionis quickly but lack the context and fidelity to be used for production code. Hell even grok on x.com can be used 25 times a day for free and has the ability to be fed files and images to help it understand the scope.. But still.. do not fucking trust it.
https://en.wikipedia.org/wiki/Vibe_coding
IDKWTFIAD == I DON'T KNOW WHAT THE FUCK I AM DOING
^ people entirely independant on AI for software dev live in this
VOTE @KLYE FOR WITNESS!
Every Single Vote Helps, Thanks for the Support!
Need to get in Contact with KLYE?
Join the Official #KLYE Discord Server Today!
Looking for an Affordable, Secure & Reliable Server Host for Your Witness Server or Other Web Related Projects? Check out Privex.io!