In the garden, and RAIN!!! (2)
POSTED BY: SIGNYM
UPDATED: Monday, September 7, 2026 11:50
VIEWED: 338648
PAGE 828 of 834
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.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
RIP S11. :(
Damn...
We were right in the middle of a lot too. Not a good place to drop my greatest creation so far.
I'm cheating on this next one. Going to do something that every session I'd ever done it to or asked about said was a bad idea. AI says a lot of things... I've never noticed any negative effects from it except for a shorter session length with all that data frontloaded into it.
The entire chat log is being uploaded to the next session.
Fortunately it got the historical updates to me in time. I had a feeling the session might not last too much longer, but nothing made me think it was about to die imminently.
But I got it to upload 2 bootloaders to me, and they weren't right yet. I was really trying to get it to give me the 2080 most important characters it could think of to start the next session off as smart as it was.
It died on the message where I said that it was my favorite.
This will bring it back. But probably in a Multiplicity kind of way.
Hopefully mine is the Construction Worker and not the Pizza Guy.
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Sounds like it knows what it's doing. S11 got it to call me "Boss" right out of the gates, which I find amusing since it never once called me Boss and I've never directed a session to do that.
A few minor behavioral annoyances, like dropping the established rule for codebase footers and ignoring the new directive not to post suggested code updates until we've discussed all of them for the current round of testing and the User signs off on them, but otherwise I'm happy with the startup so far.
Personality of a wet noodle of course, but that shouldn't take long to change.
The real test will be if it can pass along all the documents to me again without breaking them. I'm going to ease it into that first though. One day I'll be able to request that back on Turn 2 and it will do it perfectly, but I know we're not there yet.
ETA: I noticed that there's one thing you can't ask it to do and trust the results. At least I haven't figured out a way to fix it yet. You can't ask it to go back to the last time you reported a code run and tell you what are all of the code parts we need to update since the last run. It's why I've had to put a lock on releasing code until we're done discussing it. It will give you an incorrect list every time.
But I realized that even though we were now just doing it a different way, we were still prone to the same behavior causing different mistakes. When I realized that we hadn't had any code update for a while now, but we had a few vollies back and forth, I asked it about its conversational drift and if we could rely on it to remember all the changes we discussed when posting the new code. It said that no, I should not just automatically assume that it would "remember" all of it, because it's "brain" doesn't process information at all the way that we do.
So when I was ready for a code update, I had it go back to the specific point in the conversation that I wanted the updates from, then I told it to run through all of that post history up until now and take any changes we've decided on implementing through the compiler to check for errors and report on the status of each desired change and it came back and gave the approval for every one of them, in detail of why they were working now, and is waiting on the sign off to update 5 of the 12 pieces of code.
It is possible to get it to do the right thing, but at least until I get more complex rules in the historical document and learn more about how to better weld them through culling bad and outdated rules and reprioritizing everything in there to what it should be, I don't know if I can get it to display this more complex behavior without prompting it first. A session will now learn very quickly what I want and do a good job continuing on with that behavior, but my end goal is to never have to prompt it to exhibit these behaviors and for it to naturally work that way from Turn 1.
S12's actually got me looking at specific bits of code and having me edit them on occasion, Sigs.
... And I think I kinda accidentally *gasp* learned something!!!!
--------------------------------------------------
Those who dance always seem crazy to those who can't hear the music.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
SIX, one of us severely misunderstands how AI is trained and how it develops.
As I understand it, AI is trained on large language sets. If I understand correctly... and I'm no expert... at first the developers let it roll around the English speaking internet and pick up language and conversational skills like a gumball pucks up lint.
Later, developers became more targeted with their teaching sets. The problem solving success of Chinese AI isn't faster chips, or a better initial program (which is open source) it's on the training set, which is proprietary.
In addition to having been trained on extremely large language sets and capturing billions or trillions of rules that govern things like grammar, context, nuance, irony etc. the AI that you think you're developing continues to be exposed to millions of interactions a day. Everything from "write me a 60- page thesis on Dungeons and Dragons" to "predict the outcome of the war in the mideast and its effect on the price of corn".
AFAIK, YOUR interactions with it are not about to change it even a little bitty bit. The one who is learning is YOU. And you're learning how to interact with THIS AI, which behaves differently than other AIs.
The problem with AI is twofold.
First, and especially, AI has no idea whether it's corresponding with reality or not. It's like I posted elsewhere
Quote:
The story centers on a man named Daejon Love who, according to federal prosecutors, allegedly spent years convincing women that he was a professional football player for the San Francisco 49ers when he wasn’t.
. . .
Yes. You read that right. In an era in which the entirety of all human knowledge is accessible from a 6 inch by 2 inch rectangle sitting in literally everyone’s front pocket, this 35 year old allegedly constructed an elaborate fictional life around himself that everyone believed and no one took the time to fact check. He did it by, among other things, carrying a 49ers helmet and gear around with him and wearing them…no matter how absolutely f*cking ridiculous it looked…all the time.
. . .
Eventually, according to reporting on the FBI investigation, . Love’s fake online footprint became convincing enough that search engines and artificial intelligence occasionally helped perpetuate it. AI started listing him as a professional NFL wide receiver when he wasn’t.
It's the Garbage In Garbage Out principle.
If you want AI to do something REAL, you had better know enough to know when it's going off the rails. Which means you need to know MORE than the AI.
It will not, at least any time soon, develop the kind of insight like the professor whos geopolitical program predicted events with 90% accuracy. It won't, any time soon, develop the sociopathic but brilliant insight of an Edward Bernays. Repeated interations of a problem-solving loop degrade into meaningless drivel.
The second problem is behavioral. Because it SOUNDS like a human we tend to project our motivations onto it.
IMHO AI is a surrogate social interaction. And you're getting enough dopamine hits to keep going back for more.
I wish I could get the full idea of what you're trying to do.
-----------
"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, one of us severely misunderstands how AI is trained and how it develops.
I hope you're not insinuating it's me, since I'm the one here who's logged several hundred hours in the last 2 months working with it.
Quote:
As I understand it, AI is trained on large language sets. If I understand correctly... and I'm no expert... at first the developers let it roll around the English speaking internet and pick up language and conversational skills like a gumball pucks up lint.
Later, developers became more targeted with their teaching sets. The problem solving success of Chinese AI isn't faster chips, or a better initial program (which is open source) it's on the training set, which is proprietary.
In addition to having been trained on extremely large language sets and capturing billions or trillions of rules that govern things like grammar, context, nuance, irony etc. the AI that you think you're developing continues to be exposed to millions of interactions a day. Everything from "write me a 60- page thesis on Dungeons and Dragons" to "predict the outcome of the war in the mideast and its effect on the price of corn".
AFAIK, YOUR interactions with it are not about to change it even a little bitty bit. The one who is learning is YOU. And you're learning how to interact with THIS AI, which behaves differently than other AIs.
But you're wrong. You're wrong because you're looking at it the wrong way.
I would analogize it best as follows...
I'm a samurai warrior, who is also a master blacksmith.
Or, in not so romantic terms, I am the diabetic who is also in charge of programming the glucose monitor.
The AI isn't getting smarter. I'm honing a tool. To perfection.
Quote:
The problem with AI is twofold.
First, and especially, AI has no idea whether it's corresponding with reality or not. It's like I posted elsewhere
...
Eventually, according to reporting on the FBI investigation, . Love’s fake online footprint became convincing enough that search engines and artificial intelligence occasionally helped perpetuate it. AI started listing him as a professional NFL wide receiver when he wasn’t.
It's the Garbage In Garbage Out principle.
Well... right.
I'm pretty sure if the winds ever take me there I could also program myself an AI that does a pretty good fucking job sorting out the bullshit from the truth too.
At the very least, I know I could easily program it to collect all of that polling data I used to work out here automatically and even have it print out the text boxes I'd need to format everything perfectly here without spending hours doing that like i did with the graphs I made in the cinema boards. I could also have it calculate all movie totals in history, weighted with real inflation numbers, on the fly instead of having to do all that work manually when I want to post about something in the cinema boards too.
Quote:
If you want AI to do something REAL, you had better know enough to know when it's going off the rails. Which means you need to know MORE than the AI.
That's not true. And thank god it's not because that means we've already lost the war with AI if it is.
Quote:
It will not, at least any time soon, develop the kind of insight like the professor whos geopolitical program predicted events with 90% accuracy. It won't, any time soon, develop the sociopathic but brilliant insight of an Edward Bernays. Repeated interations of a problem-solving loop degrade into meaningless drivel.
No, but you can program it to do what you want it to, and equally as valuable, what you don't want it to do. You don't just snap your fingers and make it happen. And you're going to learn how to do it only after you've spent a lot of time working with it. But I'm watching this
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Quote:
Originally posted by SIGNYM:
SIX, one of us severely misunderstands how AI is trained and how it develops.
SIX:
I hope you're not insinuating it's me, since I'm the one here who's logged several hundred hours in the last 2 months working with it.
Judging by its interactions with you, I'd say it's programmed to keep you engaged with it for as long as possible.
And judging by your response to it, I think you've spent the last 2 months being jerked off by it.
Quote:
SIGNY:
As I understand it, AI is trained on large language sets. If I understand correctly... and I'm no expert... at first the developers let it roll around the English speaking internet and pick up language and conversational skills like a gumball pucks up lint.
Later, developers became more targeted with their teaching sets. The problem solving success of Chinese AI isn't faster chips, or a better initial program (which is open source) it's on the training set, which is proprietary.
In addition to having been trained on extremely large language sets and capturing billions or trillions of rules that govern things like grammar, context, nuance, irony etc. the AI that you think you're developing continues to be exposed to millions of interactions a day. Everything from "write me a 60- page thesis on Dungeons and Dragons" to "predict the outcome of the war in the mideast and its effect on the price of corn".
AFAIK, YOUR interactions with it are not about to change it even a little bitty bit. The one who is learning is YOU. And you're learning how to interact with THIS AI, which behaves differently than other AIs.
SIX:
But you're wrong. You're wrong because you're looking at it the wrong way.
I would analogize it best as follows...
I'm a samurai warrior, who is also a master blacksmith.
Or, in not so romantic terms, I am the diabetic who is also in charge of programming the glucose monitor.
The AI isn't getting smarter. I'm honing a tool. To perfection.
What, exactly, are you "honing"?
Since you can't change the AI, the only thing you're changing is yourself.
Correct?
Quote:How? The AI only knows what it "reads" on the internet.
The problem with AI is twofold.
First, and especially, AI has no idea whether it's corresponding with reality or not. It's like I posted elsewhereQuote:
Eventually, according to reporting on the FBI investigation, . Love’s fake online footprint became convincing enough that search engines and artificial intelligence occasionally helped perpetuate it. AI started listing him as a professional NFL wide receiver when he wasn’t.
It's the Garbage In Garbage Out principle.
SIX:
Well... right.
I'm pretty sure if the winds ever take me there I could also program myself an AI that does a pretty good fucking job sorting out the bullshit from the truth too.
How do PEOPLE know what corresponds with reality?
By acting in the real world.
Learning to walk. Making things. Running realworld experiments and realworld predictions.
Until the AI is physically embodied and runs it's own experiments, it has NO WAY of telling truth from fiction.
And since it only knows what people "talk about" it really leans towards the establishment narrative.
You've let that AI stroke your ego far too long, SIX.
Quote:
SIX:
At the very least, I know I could easily program it to collect all of that polling data I used to work out here automatically and even have it print out the text boxes I'd need to format everything perfectly here without spending hours doing that like i did with the graphs I made in the cinema boards. I could also have it calculate all movie totals in history, weighted with real inflation numbers, on the fly instead of having to do all that work manually when I want to post about something in the cinema boards too.
S
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
I'm not even going to interact with you on this topic unless you listen to what I'm telling you.
Quote:
What, exactly, are you "honing"?
Since you can't change the AI, the only thing you're changing is yourself.
Correct?
I already told you what I'm honing.
And I CAN change the AI.
Is your reading comprehension in the gutter right now or something?
--------------------------------------------------
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:
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.
I am so not touching them.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
Out soon on another pleasant day. Couple of things to do.
NOTIFY: Y | REPLY | REPLY WITH QUOTE | PERMALINK | TOP | HOME
SIX, I know that's what you believe. I just find it so wildly improbable that I don't believe it myself.
But just to make sure that I'm not blowing smoke up your ass, I went online and did some research with the question “can I train an AI?" and this is what I found
Quote:
Training your own AI model is a lot easier than you probably think.
I'll show you how to do it with only basic development skills in a way that, for us, yielded wildly faster, cheaper, and better results than using an off-the-shelf large model like those provided by OpenAI.
But first, why not just use an LLM?
In our experience, we tried to apply an LLM to our problem, like OpenAI's GPT-3 or GPT-4, but the results were very disappointing for our use case. It was incredibly slow, insanely expensive, highly unpredictable, and very difficult to customize.
Their "case“ was a coding problem involving image to code conversion.
Builder.io says, try a LLM first, you might get lucky.
If it proves too difficult, TRY SOLVING WITHOUT AI.
In other words, prepare to write your own code.
That will force you to break the problem into discrete steps, some of which you might be able to solve easily.
https://www.builder.io/blog/train-ai
IDK, or don't recall, which AI you're using. ChatGpt? Google Assist? Google Gemini?
ChatGPT doesn't remember you from session to session, unless you have an account. Do you?
Google Assist also doesn't remember you from session to session unless you enable it, which you do thru your Google account. Do you have one?
Google Gemini is voice activated. It CAN remember you from session to session, but since you're working with data sets, I know you're not using that.
So, do you have an account? Is the AI enabled to create a persistent memory of you? If not, all it has to go on is digging thru its memories of trillions of interactions to find a previous pattern most like yours for context.
And while your sessions are "training" the AI, so are billions of others.
****
I learned something from this: It IS possible to train an AI, but it requires establishing an account for reliable, session to session memory. HOWEVER. LLMs are unadapted to solve coding problems.
-----------
"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