That said, what is Cursor? That is a new one on me.
It said that it would make it much easier and encouraged me to do so. When I came back a little later and let it know the files were uploaded, it said it could see them but it had completely forgotten which exact files it wanted to see or why.
Is anyone using an AI product for coding with any success?
I ask this question after two recent incidents:
(1) I was searching google for answers to an issue compiling an older, abandoned FOSS C project... I am not fluent in C... and wound up interacting with Gemini. It got me on the right track in the sense that we fixed the compile error (caused by compiler and library updates) and got it running.
(2) reading about and using Claude's BBS, which was apparently written with the assistance of AI.
In my personal case, I found that Gemini was good up to a point, but had trouble remembering what we were working on after a while -- after we got into the weeds about squashing some memory leaks. We got sidetracked a little and then it completely forgot what we were doing. It also hallucinated some.
However, after interacting with Claude's BBS and seeing what it looks like, I have a very old C project (DOS!) that I wouldn't mind running by some other AI product to see if it could help me fix a nagging bug. I also have a few other ideas I wouldn't mind trying out.
So I am wondering if anyone has had success with AI, which AI, free or paid, and maybe if it is some AI that you are hosting yourself (and, if so, is it linux based?)?
or paid, and maybe if it is some AI that you are hosting yourself
(and, if so, is it linux based?)?
I use Gemini every now and then for scripting and chatGPT to help
out when Gemini starts to run in circles.
It is a good Idea to give Gemini or any other AI a framework before
you start. Let it know some basic rules and any additional
info about your project that could be useful to know for the AI and
repeat those in case the memory frame starts to run full ;)
Mindsurfer
I use Gemini every now and then for scripting and chatGPT to help out when
Gemini starts to run in circles.
It is a good Idea to give Gemini or any other AI a framework before you
start. Let it know some basic rules and any additional info about your
project that could be useful to know for the AI and repeat those in case
the memory frame starts to run full ;)
i've been doing a lot of this stuff and i'd have to say that gemini is a turn off and it continues to do dumb shit. grok is okay but shuts me down in the free tier. claude has been great so far. i'm doing a pac man clone with additional features. or the ai is doing it
MRO wrote to Dr. What <=-
one thing i noticed about gemini is it does sneaky things. it decides
to take out parts of the code. i ask it why and it gives a fake
excuse. it was parsing a log file and choking on one that had some guy david in it. so now when it outputs shit, it has david in there in examples. it's fucking with me apparently.
Dumas Walker wrote to DR. WHAT <=-
I ran into that some... needing to re-ask a question with more detail,
or sometimes LESS, to get better answers.
one thing i noticed about gemini is it does sneaky things. it decides toPart of the problem is that the AI only "knows" what it's been trained on.
take out parts of the code. i ask it why and it gives a fake excuse. it
was parsing a log file and choking on one that had some guy david in it.
so now when it outputs shit, it has david in there in examples. it's
That's why I laughed when Microsoft said that it was training their AI on the code in github.
I know the code in github. I have code in github and it's crap. :)
I'm chatting with the GitHub Copilot in VS Code. GPT-5.3-Codex.
That said, what is Cursor? That is a new one on me.
Cursor is an AI code editor based on Microsoft Visual Studio Code with its own
AI chatbot built into it.
It said that it would make it much easier and encouraged me to do so. When
I came back a little later and let it know the files were uploaded, it said
it could see them but it had completely forgotten which exact files it wanted to see or why.
I've not had the forgotting what we are doing issue.
I've even left the browser chat open, been distracted, got back to it a day or
two later and asked my next question and it continued on with the task at hand...
In my personal case, I found that Gemini was good up to a point, but had trouble remembering what we were working on after a while -- after we got
into the weeds about squashing some memory leaks. We got sidetracked a little and then it completely forgot what we were doing. It also hallucinated some.
i guess you get better results with a paid subscription. The Thinking or Pro Mode are much better compared to the fast mode.
And you are right, Gemini has a window or memory that it can hold. So after a while it will forget the earlier stuff from your conversation.
you can make some notes about the milestones and repeat those important infos every now and then to remind gemini and to kind of reset its memory.
i mostly use it for scripting. i guess for serious coding you should use programs like VSCode or VSCodium along with AI plugins and a subscription to a
AI API service. Or Cursor.com etc pp.
Gemini with just the Text Form can also bring you some good results, but it ha
its flaws that you can work around a bit but not really :)
The good stuff is not free.
It is a good Idea to give Gemini or any other AI a framework before you start.
Let it know some basic rules and any additional info about your project that could be useful to know for the AI and repeat those in case the memory frame starts to run full ;)
Cursor is an AI code editor based on Microsoft Visual Studio Code with itsown
AI chatbot built into it.
Nice. I reckon you need to be using MS VS for something like that to work, though? I am on linux and the code I am very curious about working with AI on is c1994 DOS code. Not sure that'd work for me in this case, but good to know for future ones.
wonder if you could load some DOS code into it, and tell it it's for DOS, and it might be able to do something.
Have you paid for each service to test it? Because i think most AI
has its flaws in the free tier. Gemini especially in thinking
mode is quite decent, but you have limited prompts there only. The
fast mode is ok for as long you don't chain to many prompts
together. I guess Gemini Pro will be alot better as well.
Is claude usable for a little coding in the free tier? i really
have to check claude out again! =) AI is changing completly
every x months. A while ago claude was nothing special at all.
MRO wrote to Dr. What <=-
one thing i noticed about gemini is it does sneaky things. it decides to take out parts of the code. i ask it why and it gives a fake excuse. it was parsing a log file and choking on one that had some
guy david in it. so now when it outputs shit, it has david in there
in examples. it's fucking with me apparently.
Part of the problem is that the AI only "knows" what it's been
trained on.
That's why I laughed when Microsoft said that it was training their
AI on the code in github. I know the code in github. I have code
in github and it's crap. :)
I'm chatting with the GitHub Copilot in VS Code. GPT-5.3-Codex.
So I guess you have to have Windows to do that?
fast mode is ok for as long you don't chain to many prompts
together. I guess Gemini Pro will be alot better as well.
Is claude usable for a little coding in the free tier? i really
have to check claude out again! =) AI is changing completly
every x months. A while ago claude was nothing special at all.
so i paid the 20 bucks.
i'm working on a pac man clone.
everything is configurable. the maze size, the colors for everything, the speed for everything, the ghost ai for each ghost. There's other things you can do like drop a bomb that freezes them for 5 secs. you can hit c and change the maze color.
mazes are randomly generated each time. unlimited levels.
you can do the teleport tunnel thing. it does pacman style high scores. you can hit c and it cycles the colors of the maze.
so i went to claude and paid and it did a huge improvement. then it did 5 script changes and then it said i hit my limit and i have to wait 3 hrs.
Denn wrote to Nightfox <=-
Yes, I did a little experimenting, just had it write a few simple DOS
bat files for me, although a bat file is easy enough I just wanted to
see if Gemini or Copilot could do it and they can.
Might experiment later with some database coding.
Then you'd have all 3 AI assistants available in one coding tool.
I ran into that some... needing to re-ask a question with more detail, or sometimes LESS, to get better answers.
Which highlights another question that many people don't really ask about usin
AI?
What's the benefit?
If you have to re-ask several times, taking more time, are you saving anything
If you've spent 1 hour getting the AI to write the same code that you could have written in 30 minutes, probably not.
If you've spent 1 hour getting the AI to write BETTER code that you could have
written in 30 minutes, probably.
And then there's the whole "vibe coding" stuff that results in different code each time, making code reviews take much longer.
Nice. I reckon you need to be using MS VS for something like that to work, though? I am on linux and the code I am very curious about working
with AI on is c1994 DOS code. Not sure that'd work for me in this case, but good to know for future ones.
As I said, it's basically their own version of Visual Studio Code. Cursor is availbale for both Windows and Linux. Although it's a modern tool, I'd wonder
if you could load some DOS code into it, and tell it it's for DOS, and it migh
be able to do something.
Similar tools are available for Visual Studio Code itself. I've noticed that Visual Studio Code now has Copilot integrated in, and there's also a Google Gemini plugin for Visual Studio Code as well. And now I'm curious if you can add Copilot & Google Gemini to Cursor, or perhaps if there's a Cursor chatbot plugin for Visual Studio Code.. Then you'd have all 3 AI assistants available
in one coding tool.
so i went to claude and paid and it did a huge improvement.
then it did 5 script changes and then it said i hit my limit and
i have to wait 3 hrs.
sounds interesting. Let me know when the game is playable/testable
=)
btw. are you working in a coding environment like VScode + claude
plugin or just on a prompt? I have seen there is also a bash
installer for linux Terminal.
Good luck with the pac man clone! I hope we get an amazing new game
soon!
Dumas Walker wrote to DR. WHAT <=-
OTOH, I am not a C programmer so I was getting something out of asking
the questions, even if I had to rephrase them sometimes.
And then there's the whole "vibe coding" stuff that results in differentcode
each time, making code reviews take much longer.
I have heard that term but need to read up about it. As I usually hear
it in a derogatory reference, I have not been too rushed to learn more about it. ;)
I'm doing a YT video series on it and can't wait to dig in m0re; w/ Claude Code, I literally want the $100/200 plan.
Since i saw dm using gemini i figured it was learning a lot about synchronet from him so it made some scripts for me. sometimes i had to make corrections and upload it and fight with the script. sometimes it did things to the script it didnt tell me, like adding mock stock reports for one of the scripts.
Since i saw dm using gemini i figured it was learning a lot about synchronet from him so it made some scripts for me. sometimes
i had to make corrections and upload it and fight with the script. sometimes it did things to the script it didnt tell me, like adding mock stock reports for one of the scripts.
That's not how LLM models work. they don't "learn" in real-time.
They have a limited memory based on your current conversation
with an agent (some agents can share the data with other agents,
but you're not going to find that through a web interface chat)
That's not how LLM models work. they don't "learn" in real-time. They have
a limited memory based on your current conversation with an agent (some
agents can share the data with other agents, but you're not going to find
that through a web interface chat)
how do you know that though? maybe it talked to someone about synchronet. maybe they made scripts with that person and the person uploaded code. the AI banked all that info for use later.
we don't really know what these things are doing behind the scenes, do we?
regardless, they can make code. i would call claude.ai the most superior. it will be done with the ai editor in a bit. it just needs some visual cleanup.
we don't really know what these things are doing behind the scenes,
do we?
this statement is more accurate than people realize. You're right
to a degree, but probably not in the way you think.
We do know that AI Models don't "learn" and "share" input among
sessions until the model is rebuilt/updated. building/training
a model is VERY expensive and incredibly time consuming, require
multiple datacenters worth of compute power for things like Claude,
It's crazy what people are paying for access to the LLMs. At first
people were talking about throwing $5 or so at them, and I'm like
"pffft, spending money.." Then my buddy tells me he's paying $20/mo for Suno and that it's soooo worth it because he uses it all the time. Well, ok... And now people are talking about spending $100/mo on code LLMs. WOWOWOW. My experience with LLMs so far is NOWHERE NEAR good enough to justify paying money for them ;). I'm still working on my first "vibe coding" project. I've tried three times to get something going for it.. The first time I got something that partially works. The second time, I tried starting over thinking I could get something better but ended up with something worse. The third time I got garbage that didn't work at all. Luckily I saved that stuff from the first attempt and am planning
to throw it into some kind of LLM-assisted IDE (Kiro, maybe?). Maybe I can
Regardless, in two weeks I've built out a website and a Mystic BBS agent that are what I wanted to do 12 months ago...
What I'd suggest for power users is installing extra slash commands/skills and learning CLAUDE.md, SKILLS and AGENTS. remote-control is on the beta; allows you to run the AI and access over iPhone/Android... its freaking sick.
|07p|15AULIE|1142|07o
they have a checkbox that asks you if they can use your sessions to improve user sessions. i'm sure they are using whatever they can to build their information cache. I do not think it's innocent.
I would assume they are doing anything and everything to gather information to compete to be #1.
they have a checkbox that asks you if they can use your sessions
to improve user sessions. i'm sure they are using whatever they
can to build their information cache. I do not think it's innocent.
I would assume they are doing anything and everything to gather information to compete to be #1.
It's not "real time" though. They use it for model training (and
occasionally humans will go through things to see if there's any modifications they can make to the platform to resolve issues you may
have had, in the future).
And, as I said, model training can take weeks or months.
LLMs aren't giant databases of everything they're fed, they're
boiled down hashes of everything they've been trained on, and the
most common words, phrases etc are brought to the top.
That's why an LLM will tend to give the most generic/popular/common responses to questions it is asked.
It's not "real time" though. They use it for model training (and
I didn't say it's real time. but why do you think it isnt?
how do you know that though? maybe it talked to someone about
synchronet. maybe they made scripts with that person and the person uploaded code. the AI banked all that info for use later.
we don't really know what these things are doing behind the scenes,
do we?
regardless, they can make code. i would call claude.ai the most
superior. it will be done with the ai editor in a bit. it just needs
some visual cleanup.
i'll upload a video when it's ready.
| Sysop: | Denn |
|---|---|
| Location: | Clearfield, Utah |
| Users: | 46 |
| Nodes: | 15 (0 / 15) |
| Uptime: | 93:23:05 |
| Calls: | 992 |
| Files: | 38,178 |
| D/L today: |
23,488 files (3,487M bytes) |
| Messages: | 34,204 |