In the garden, and RAIN!!! (2)
POSTED BY: SIGNYM
UPDATED: Monday, September 7, 2026 11:50
VIEWED: 338648
PAGE 829 of 834
You're still not getting it, Sigs.
No. I do not have a paid for account, as you already know full well if you've read what I've posted in the past.
EVERYTHING I am doing has been done, simply using the free-for-everyone AI assistant that is now built into a Google search bar and will, more often than not, generate a response at the top of your query results.
The way that MY PERSONAL AI learns is through the historical documents and the blueprint which dictate how it is suppose to behave (both in general and in specific situations), which is all provided to the new and very stupid AI at turn 1.
If I don't do a stellar job maintaining my own documentation so I can feed that info to a new session on Turn 1, its dumber than Ted. By a lot. And that's saying something.
I'm not training GOOGLE AI as a whole. I'm making sure that the next session that I open in Google AI has already "relearned" everything we've worked on and the rules that we've created for it when I start a new session. And that history and those rules compound on each other every session. If I wasn't doing any of that, this AI would be completely worthless to me after a session crashes because even though my code is in place, it would have no historical context for any of it and any of the established rules and protocol we've been building now over 12 sessions would be completely lost. We would just be starting from square one every time.
And, if it continues to fuck something up after a new session starts, it's either because we haven't yet figured out what EXACT rules to add to our blueprint to quarantine that undesired behavior, or it's just something we haven't gotten around to addressing yet because we're still working on other things.
If you can't grasp this concept, we can't even discuss this further because there's zero point in it.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
1) You don't have a PERSONAL AI unless you have a PERSONAL account..
2) I got that.
I've posted, over and over again, that the ONLY thing you're "honing" is your ability to reinitialize each session more efficiently. That you're learning, by stumbling around, the limits of Google Assist (eg token limits), how to (supposedly) get it to exceed its limitations, the "secret sauce" that will prompt it to do things (like send links) that it's not supposed to do.
While IMHO all experiential learning is valuable, some learning is more valuable than others.
Stumbling around for months at a time, bumbling into Google Assist's limits and trying various workarounds, is time far better spent IMHO learning something else.
Again IMHO there's going to be very little ROI unless you snag contract- style jobs from small businesses, individuals, and small institutions who would like someone to develop something for them, or research something for them, using freebie AI.
But it's your time and your dime, so have at it, and good luck.
-----------
"It may be dangerous to be America's enemy, but to be America's friend is fatal."- Henry Kissinger
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
I disagree with you wholly on your assessment of what I'm doing, but to each their own.
I'm forward thinking by design, and I can only imagine the possibilities this could lead to in the future regarding coding or even a wide variety of other things.
What I'm building here goes far beyond the code itself. I've already ingrained a host of behaviors I want my AI to exhibit on a new session that could easily be adapted to a different purpose entirely, and then build on the behavioral changes from there. We could even have sub-directories on behavioral expectations based off of what type of thing we're working on.
I'm not going to argue with you that I'm not learning a lot. I AM learning a lot here.
Just not explicitly Python code is all.
In any event, we finally built the new and improved codebase that has been stripped almost entirely of white/blacklist legacy bullshit code and has been rebuilt from the ground up to rely on our own function library, innate commands available to the python compiler and a much smaller codebase footprint overall.
Not only that, but an execution run on the largest of the 3 current testing environments used to take around 5 minute and the hard drive CHUGGED the entire time. Now the drive hardly makes a sound and the entire process is finished in under a minute.
Oh, and bottom line, Sigs, though I really hate to admit it to myself, with just a minor tweak here or there depending on the situation, I've just "invented" something with this AI that, if time travel were ever invented, could have given me anywhere from 5 to 7 years of my waking hours in this lifetime back to me that was spent on manually doing things that I would never have to do again for the rest of my life.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Back and done for today.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
SIX, "you" don't "have" an AI. You share it with hundreds of millions of users. To the extent that it" remembers" your project from one session to the next apparently depends on whether or not it can pick you out of the billons of sessions it's stored.
Now, IDK whether this will work, it's a total guess on my part, but you might make that memory more persistent by giving yourself an indentfier at the end of each session. eg. "I am (fill in the blank, with lots of alphanumeric and special characters)
I want you to remember (tell it what you want it to remember)"
It may allow more persistent and reliable memory of what it's been doing and how you approach things.
And at the beginning of the next session, tell it to look for the most recent session with that identifier, and follow the last instruction. Like I said, total guess on my part. It may bot work, or it may be more trouble than it's worth.
Oh, and BTW, it's great that you're learning.
-----------
"It may be dangerous to be America's enemy, but to be America's friend is fatal."- Henry Kissinger
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Quote:
Originally posted by SIGNYM:
SIX, "you" don't "have" an AI. You share it with hundreds of millions of users. To the extent that it" remembers" your project from one session to the next apparently depends on whether or not it can pick you out of the billons of sessions it's stored.
Please stop.
I understand that. I'm not an idiot. And I'm not at all delusional about what exactly it is I am doing here.
Quote:
Now, IDK whether this will work, it's a total guess on my part, but you might make that memory more persistent by giving yourself an indentfier at the end of each session. eg. "I am (fill in the blank, with lots of alphanumeric and special characters)
I could do that, but that's just filler. S11 apparently got S12 to refer to me as "boss" with something it put in the blueprint. I don't really like it, and I will probably find that part of the blueprint and delete it when going forward so it doesn't do that. I've never been anybody's Boss, and I don't want to be. Not even when playing pretend with an AI.
Hell... I could very easily make it refer to me God on the next session if I wanted it to.
Who knows though? Maybe there is some sort of "secret" ID that I could add to a session that would remember our exact history where Google record it somewhere. And instead of uploading 19,000 to 32,000 lines of previous chat history for every single session, I could just start pointing it to Google chat ID serial numbers and use that as a future reference point. That would be a HUGE upgrade to the process if it were possible.
It would actually surprise me more to know that Google wasn't cataloging and warehousing every single interaction their AI had with people, somewhere. And if not EVERY one, they might have filters in place to save any that they consider even moderately interesting, so they're saving all of the interactions that go just beyond maybe the first 3 prompts back and forth to filter out all the basic bullshit.
I could try to see if I could get it to accidentally figure out this identifying number that a new session would not have any idea about. But I honestly think that unless they fucked up, Google wouldn't allow the AI to ever have knowledge about that even being a thing that happens, and certainly shouldn't allow end users like me to exploit that. I would assume that sort of cataloging would happen far outside the boundaries of the AI sessions themselves, and is likley handled by another system that rounds up thousands of completed sessions at one time and does it all after they're terminated.
Quote:
I want you to remember (tell it what you want it to remember)"
It may allow more persistent and reliable memory of what it's been doing and how you approach things.
Now you're getting it.
That's exactly what I do with our blueprint and historical document and especially the Bootloader, which is basically the PRIME DIRECTIVES with the most important behavioral adjustments to a new session.
The bootloader is only 2080 characters limit, but there is no limit at all to the size of the historical documents, or at least not one that is going to be met in years worth of doing this work. The blueprint (behavioral changes) is now 20kb long, and the thesis (historical document) that documents all changes and milestones across sessions is now around 35kb.
Even though the new session is a complete blank slate when you start it up, I can add the code, the history, the blueprint and the bootloader, which are all evolving with every session, into the first prompt and it immediately "knows" all of the pertinent history from the previous sessions and everything we've accomplished and what the goals are for the next session.
And this last one I did the "emergency" option of starting it off with the full 19,000 line chat log of the previous session since S11 died a
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
BTW>.... It's freakin' GROSS here right now. 2nd day of 4 over 90 degrees and humid as hell again.
September 1st.
What a shitty summer overall. 2 out of 10.
How's your heat, Sigs?
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Quote:
Originally posted by Brenda:Quote:
Originally posted by 6ixStringJack:Quote:
Originally posted by Brenda:
Ugh! I've got two crossed out gray cloud icons on my task bar. Have to get rid of them before they multiply.
You'll be alright, Brenda.
... Just don't feed them after midnight.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
You can touch them.
Just don't get them wet though.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Quote:
Originally posted by 6ixStringJack:Quote:
Originally posted by Brenda:Quote:
Originally posted by 6ixStringJack:Quote:
Originally posted by Brenda:
Ugh! I've got two crossed out gray cloud icons on my task bar. Have to get rid of them before they multiply.
You'll be alright, Brenda.
... Just don't feed them after midnight.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
You can touch them.
Just don't get them wet though.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
Seriously dude.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Oh...
And what I thought was just a minor change this time around that helps me out when updating code parts ended up being a huge change for the process.
I've had it add a Revision # to the text header and footer of every part it updates. Not only does this make it easier for me to track these things and make sure I'm up to date on my code, but the really cool unintended side-effect of doing this is now the AI actually makes references to previous revisions when we're discussing things.
What I intended to add only as a way to visually make checking that my updates are solid easier actually opened up a doorway for the AI to structurally change the way that it handles our code updates and how it retains them in its "memory".
I didn't even ask it to do this, but now when it posts a new update it finishes off the post listing all of the parts I need to update with their corresponding revision number and when we're discussing previous changes, it cites specific revision numbers of the parts when we had zero way of doing this before.
Cool.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME