|
|
| Page 51 of 51 |
|
|
 Posted: Fri, 21st Aug 2026 17:20 Post subject: |
|
 |
| Quote: | Goldman Sachs research reveals that Al is eliminating approximately
16,000 U.S. jobs monthly
|
Enough yet, nope. |
|
| Back to top |
|
|
LeoNatan
☢ NFOHump Despot ☢
Posts: 75248
Location: Israel
|
| Posted: Fri, 21st Aug 2026 18:55 Post subject: |
|
|
Here is the decisive underpinning: AI is not LLM.
|
|
| Back to top |
|
|
vurt
Posts: 14509
Location: Sweden
|
| Posted: Sat, 22nd Aug 2026 01:29 Post subject: |
|
|
| Il_Padrino wrote: |
I still wouldn't trust it to generate all the code itself, though. It models it's syntax to the examples you feed it (like old code) and if you don't ask it whether it's actually good code, it just goes along.
But building the app part by part seems the perfect balance. I frequently ask it to audit the entire solution, which seems to help a lot too. The code it suggests at this point I can pretty much copy-paste into my solution. |
Glad you having fun!
Pretty sure it does a good job at generating the code, but i use the most expensive models for planning it, then maybe a less expensive one to do the work. When auditing i rarely run into any model saying it's bad, in fact it often gets praised.
This was absolutely not the case with my first game, the models would find bugs, dead code, bad practices. This happens very rarely now, but yes you should do an audit from time to time.
Are you working inside Claude Code? If you are working from the chat window in the browser then you are missing out a lot. It's how i started, but it doesn't come close to how it is working inside Claude. You have a lot of extensions you can use too like Ponytail and others.
Token savers can be good to, for example
https://github.com/rtk-ai/rtk
https://github.com/gglucass/headroom-desktop
https://github.com/samuelfaj/distill
https://github.com/chopratejas/headroom
|
|
| Back to top |
|
|
DXWarlock
VIP Member
Posts: 12279
Location: Florida, USA
|
| Posted: Sat, 22nd Aug 2026 15:57 Post subject: |
|
|
| vurt wrote: | been quite amazed with how good chatgpt SOL is though + 2 resets in like 2 days which is great. so many funny and woah moments while working on a new game.
I had an image file which i was going to implement in a certain way, just called "lower.png" when it was doing something else it found that image, looked at it (its just the lower part of a bulkhead, kind of long steel part) and it just implemented it straight away without asking how i wanted it, and it did it exactly how i had intended it which is kind of complex.
I was just about to create long prompt for how to do it, not needed, that kind of stunned me, like "are you a fucking mind reader?" |
Do you run into the double edged sword of that too?
For me the times it auto did something I want, is outweighed by the times it decided to do something I didn't want.
---
Like debugging a comfyUI log file. I asked it to read it and see if it could explain the obtuse cuda errors causing the crash. It did that, and then also decided to 'warn' me of things we talked about TWO MONTHS ago like "pinned memory is back on, as well as dynamic v-ram is disabled. (LOONG paragraph about both)" then proceeds to start trying to edit my bat file. because 2 months ago, in a separate chat I allowed it to edit it.
This wasn't even in the same chat we had months ago, a totally separate 'new' chat.
I have to stop it and tell it "I did not ask to compared it to a log from 2 months ago, I asked you to specifically explain two cuda crash errors, and the two things you flagged are user options, not issues. And you are just comparing stale data to new on them, and making that decision for me"
It then apologizes like a dog I beat with a newspaper and does what I asked to start with.
 Spoiler: | Fair. I flagged them as issues because of prior context, not because the log itself shows a problem. (...a few more lines of grovelling over doing it) |
---
Or I simply ask it: Give me a list of 50 wildcard tokens for medieval barbarian armor.
It will go and start listing things formatted like other conversation based on some unrelated, but also 'list' of things I asked for for a word document with bullet points.
While also adding color because 2 weeks ago I asked for Bard clothing color wildcards and it goes "Know what would be better if I just did? Barbarian armor...IN WILD COLORS like the bard"
----
Not a great examples, but seems claude now lately is a LOT more apt to do that. And I have to constantly tell it the equal of: That's not what I asked/wanted. Dawg you doin to much.
One of the reasons I removed its rights to write files, and only read. Despite even stating IN the project MD to always ask before any edits.
The amount of times it just started 'doing' a thing while working on a GTA5 rage plugin was outrageous. I ask like "scan the rage and LSPDFR api, is there a way to indicate which peds I injected with personality data"
It start reading, then I see the: changed file, red/green line count number, while its thinking and I have to stop it. I asked IS there a way, not for you to pick a way, start putting it in, and change a lot of code over a question "does API have that feature"
At times it feels like an over eager new hire and I went to their desk and asked: How hard it is to change the website background color it's too bland? Can a gradient be applied?
And they start slapping the keyboard and go "Not hard at all, I remember a while back we talked about colors of cars you liked and you own a cat. So I changed it to a red/blue gradient for you, and added animated gifs of cats."
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
|
|
| Back to top |
|
|
Il_Padrino
Posts: 8118
Location: Greece by the North Sea
|
| Posted: Sat, 22nd Aug 2026 16:17 Post subject: |
|
|
| vurt wrote: | | Il_Padrino wrote: |
I still wouldn't trust it to generate all the code itself, though. It models it's syntax to the examples you feed it (like old code) and if you don't ask it whether it's actually good code, it just goes along.
But building the app part by part seems the perfect balance. I frequently ask it to audit the entire solution, which seems to help a lot too. The code it suggests at this point I can pretty much copy-paste into my solution. |
Glad you having fun!
Pretty sure it does a good job at generating the code, but i use the most expensive models for planning it, then maybe a less expensive one to do the work. When auditing i rarely run into any model saying it's bad, in fact it often gets praised.
This was absolutely not the case with my first game, the models would find bugs, dead code, bad practices. This happens very rarely now, but yes you should do an audit from time to time.
Are you working inside Claude Code? If you are working from the chat window in the browser then you are missing out a lot. It's how i started, but it doesn't come close to how it is working inside Claude. You have a lot of extensions you can use too like Ponytail and others.
Token savers can be good to, for example
https://github.com/rtk-ai/rtk
https://github.com/gglucass/headroom-desktop
https://github.com/samuelfaj/distill
https://github.com/chopratejas/headroom |
I'm clearly in the honeymoon phase 
But yesterday I asked it to clean up a few duplicate images from an old library folder, and it warned me that they were duplicate for a reason, something I had completely forgotten about. I never explained it the mechanics, it was able to derive that from just the filenames. Saved me a few hours restoring those imgs if I had just deleted them myself
So I don't mind it being proactive, and the stuff we discussed are maintained automatically in project docu. And it's not glazing me either, I asked it for its honest opinion, and it then (albeit gently) completely broke apart my old code
But I have to ask pretty specific questions, like "the app is doing it like this, but is this actually the best solution or are the better alternatives?".
Not using Claude Code yet, just the collaboration chats grouped together inside a project (and I'm using the pc app, not the browser).
what are the advantages of using Code? Unit testing and such? I won't let it write/modify my code anyway.
The numbers don't decide, the system is a lie. The river running dry, the wings of butterflies.
And you may pour us away like soup. Like we're pretty broken flowers.
We'll take back what is ours. One day at a time.
|
|
| Back to top |
|
|
DXWarlock
VIP Member
Posts: 12279
Location: Florida, USA
|
| Posted: Sat, 22nd Aug 2026 16:36 Post subject: |
|
|
| Il_Padrino wrote: |
Not using Claude Code yet, just the collaboration chats grouped together inside a project (and I'm using the pc app, not the browser).
what are the advantages of using Code? Unit testing and such? I won't let it write/modify my code anyway. |
Code can run in a folder (say your root repo folder), with its own .claude subfolder. SORT of like a project, but different. Think of it like github, where you have your github and each repo under it.
Claude Code can run in the root of your 'repo' folder you point it to. So can have like 40 seperate 'things' you are working on. And one global place for rules, .MD instructions for global guidelines and 'rules/schema/best practices' etc. It walks up the folder tree looking for CLAUDE.md files, so a top-level one at your root applies to everything under it, and each project can have its own on top of that.
Plus the parts chat straight up cannot do: it can do compiles, direct edits, leave context notes for itself in MD files, git integration. It reads your whole codebase off disk. No more pasting files or hoping the project upload grabbed the right ones. It greps across everything, follows imports, finds call sites on its own. And when you do let it act, it writes a change, runs the tests, reads the failure, adjusts, re-runs. In chat you're the middleman on every step.
This is a mix of good AND bad, depending on your viwepoint.
Other bits worth knowing... CLAUDE.md at the repo root gives you persistent instructions that load every session so you stop re-explaining your conventions. It runs any shell command and reads the output itself, linters, builds, curl, whatever.
It'll also grab whatever it needs to get the job done. Missing a compiler, some npm package, a CLI tool, a zip from some release page? It'll pull it down and install it. Basically anything you'd do from a terminal to set up an environment, it just does.
Skills, subagents, hooks, slash commands are the deeper end for repeatable workflows, overkill for single file projects. But great if you have multiple files.
All good things. my complaints is mostly from I am a control freak on my code. If you are working with me, don't even add comments to my code without me knowing  Claude LOVES it some code comments, it makes it rain with them if you let it.
One thing I let it refactor for me to see how well it did was about 1600 lines when it was done. 500 of them was comment lines...You can tell it remove them, and it will. But left to its own devices it will document EVERYTHING it changes in a comment. Even things like "Refactor changes from legacy ped call that was removed in pedstate.cs"..why add that? That function and entire .cs is gone, anyone that reads this later is just reading about things that don't exist to know what that even means.
And I dislike it trying to change even one line (even if it was roughly what I was going to do) without asking me, or if I didn't explicitly tell it to. Even if it finds a ID10T error I made, I don't want it touching it to fix it, unless I can see how its fixing it and it asks first. Some people prefer the ease of it auto fixing mistakes with its own best judgement.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Last edited by DXWarlock on Sat, 22nd Aug 2026 16:46; edited 1 time in total
|
|
| Back to top |
|
|
Il_Padrino
Posts: 8118
Location: Greece by the North Sea
|
|
| Back to top |
|
|
DXWarlock
VIP Member
Posts: 12279
Location: Florida, USA
|
| Posted: Sat, 22nd Aug 2026 16:54 Post subject: |
|
|
I'm with you on that.
I switched because the good options, slightly outweighs the bad ones. And you can tweak it to alleviate most the bad ones (which is different per person on what is 'bad').
It's just a 'starting from square one' type thing on teaching it again. You start in there, and start narrowing down: This is good, do this. This is BAD, very bad. Do not ever do that again.
I do enjoy that I can edit my code, tell it 'make it so' and it will compile my release DLL with packed and formatted ini config files ready to go.
I write the plugins in C# myself..mostly. using Claude as a "I can't be assed to learn a whole API environment for TWO game overhauls, you are my lookup tool when I need a thing".
But releasing it normally that means opening Visual Studio, loading the csproj, making sure it's pointing at the right Rage and LSPDFR reference DLLs (AND right version of them and I have the newest to reference), running a Release build, then grabbing the output from bin\Release.
I just tell Code "build me a release" and it runs the compile against all the references itself, checks the .NET target matches what Rage/LSPDFR are expecting, warns me if something's off, and hands me back a zipped DLL with formatted ini config files for it, ready to drop in.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
Last edited by DXWarlock on Sat, 22nd Aug 2026 16:58; edited 4 times in total
|
|
| Back to top |
|
|
|
|
| Posted: Sat, 22nd Aug 2026 16:55 Post subject: |
|
|
if these LLMs keep scraping all websites that rely on adds like reddit insta fb etcetera ... wont they just simply start blocking the llm scrapers or try to add the scrape with adds ?
i have a feeling this is peak add free search results right now and future will be massive enshitifivation or pay to search
|
|
| Back to top |
|
|
DXWarlock
VIP Member
Posts: 12279
Location: Florida, USA
|
| Posted: Sat, 22nd Aug 2026 17:02 Post subject: |
|
|
@PickupArtist
Dunno if it goes that route, but SEO poisoning has been a thing waaaay before LLM.
If they found ways around it since the time of AOL. They will here.
It's a constant cat and mouse battle. LLM isn't a different game. Just a different mouse.
-We don't control what happens to us in life, but we control how we respond to what happens in life.
-Hard times create strong men, strong men create good times, good times create weak men, and weak men create hard times. -G. Michael Hopf
Disclaimer: Post made by me are of my own creation. A delusional mind relayed in text form.
|
|
| Back to top |
|
|
vurt
Posts: 14509
Location: Sweden
|
| Posted: Sat, 22nd Aug 2026 21:08 Post subject: |
|
|
the scarping must be done intelligently or it would be so full of crap, it likely already is but...
i totally get why they are buying whole libraries of books..
|
|
| Back to top |
|
|
| Page 51 of 51 |
All times are GMT + 1 Hour |
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
Powered by phpBB 2.0.8 © 2001, 2002 phpBB Group
|
|
 |
|
It knows my naming convention pretty well now, but occasionally it gets something wrong so I'm fine with copy pasting and manually validating everything for now.
