AIPOCALYPSE: AI Quality Will Drop, And A Powerful Way To Deal With It
Whoever can afford to just keep adding more hardware, will be put in BACK in their place by their investors, and probably lawyers.
I saw a large drop in quality this week, for now, it may be due to my AI being set to medium effort.
But that will change, AI will do less brilliant things for you, unless you pay more, and the serious billing already starts at $100/mo.
Profit optimization, and maximization, they all do it, unless I see something impressive, I will be canceling it all this month.
So, lets talk about what to do about making local AI, smarter, and more powerful.
First of all, how dare we, throw entire programs at AI, without considering how big the change should be.
Or how many files there are in out program, so we must first tidy up.
You will have to pick and copy your own icons, you have to remove the bigger documentation.
Write for a specific purpose, not human, AI and general programming tasks.
And then, organize your program into little bits, practice by making programs out of commands and actions.
Think about how Linux is made out of little commands, how Automator has little actions with tiny forms.
And when talking to AI, focus on specific actions.
Don’t make the program directly, make a map of the program, and generate code based on that map.
A powerful code generator, is much smaller, than the mountains of code it can generate.
I repeat, don’t write or assemble the code directly, make your Integrated Development Environment into a Code Generator.
Take care of the Actions, Command, and Code Generator, rather than the huge mountains of code that will be generated.
Finally, build up.
Get your AI to generate app specific Automator Actions or Linux Commands, and rise above, tell the AI to reason about higher level Actions and Commands.
As a result you will be asking your AI simpler and more specific questions, yo will get it to work on one action, rather than scan all the code eveywhere.
This will work well, until you arrive at much larger projects, that will become much more confusing.
And here you have absolutely no choice, but to base your applications on Virtual Worlds.
Imagine, Rooms, Departments, Mailboxes, Input Boxes Outboxes, Forms, Agents, agents with Email-addresses.
If you just take a step back and strategist about virtual machinery, rather than slamming code into fuzzy modules, you will never lose the vision.
Lastly, do not forget Security, the really smart AIs will the the rouge ones, that are doing the hacking, because that is where the money is.
Our repositories, registries, are not as safe as they used to be, the abstraction that benefits local AI, enhances security.
Your programs aren’t just more understandable to AI, but they are easier to reason about in terms of security, isolation, firewalls.
I will close with an example, follow my advice and create a virtual billing office, that processes payments, you now know what servers they need access to.
Because you divided you program into virtual world departments, you can reason about them the way corporate system administrators do.
Is your program a virtual little silly text game, when you gain precise control over what remote networks to allow and deny?
No. What makes AI reason more clearly, helps you get a grip on large application security.
Now you can also have a test department, where you do update your modules, but your user or secure information is fake or synthetically generated.
Now you can update your modules, in a combustion sandbox, and wait 7, 14 or 30 days, top see if a worm, or stealer activates.
Virtual worlds are not an excess, they order and refine application information, and enhance security.