Talk is cheap. Your point would be best proved by a release. When do you expect that to happen?
2 Likes
Iāve been thinking much about exactly that. I donāt have any experience with open source projects.
An early release could be beneficial for me if other would participate in the dev of the simulator.
But the biggest downside is that itās very unfinished and Iāve had negative experiences showing stuff to early (non AI related). I guess youāre just waiting for the opportunity to call my project AI slop and grilling as much as possible of it. So I will release it when I feel ready for it.
No one is forcing you to do otherwise
Weāre not, but you are talking it up and talking up the value and usefulness of the tools thatās why weāre talking about the results, how good they are and all the great projects made with it
In general youāre being pretty combative and accusatory and defensive all at once, you came here promoting the idea of using LLMs for these things, no one forced you to defend it or promote it or your project
3 Likes
That video was hilarious. They made good points.
2 Likes
Frankly, it would be unwise to allow hallucinated code to be refed to the LLM via Githubā¦ As an assembly and c++ programmer, Iām sure you remember the concept of entropy?
Better keep the bots good at their job. Right?
..right?
3 Likes
No, I just want to see the results that back up your bold claims.
However, I doubt we will see a release because Itās not possible to vibe code a decent game, let alone a flight simulator. You will give up as soon as your claude and copilot run out of low hanging fruit to pick, which will inevitably happen, if it didnāt already.
We can meet here this time next year and see if Iām right. If you manage to prove me wrong by showing a stable playable release - Iāll eat my hat and buy you a beer and three months worth of claude code tokens (if the price stays the same)
8 Likes
While weāre on this topicā¦ How come there is no class action lawsuit for back pay for free work that they now put behind a token paywall?
I donāt understand much about law, but I had the impression you can basically sue anyone to proverbial unliving using US lawfareā¦
Authors, actors, singers and graphic artists are already trying this right?
4 Likes
I think thats about accurate.
1 Like
I understand that things get heated with this topic. It was never my intention to hype or even defend anything AI. As I stated earlier, Iām mainly curious how far I can get my project with todays tools.
Among those tools is also Godot.
And no Iām surely not at the point of giving up. Not in a long shot. And I do want to try to persue this project further when AI gets to itās peak.
1 Like
Iām on your boat. But got the same realization as documented in the following thread: I have a 2nd PC running for experiments with agentic tools. Havenāt used it for some time. It gets frustrating fast. Now I use it sparely only for a auto completion workflow.
IMHO, the most important thread on this forum regarding AI. For really small projects it can be ok, quick testing some concepts or testing some ideas. But at a certain level it will break and produce garbage. You have to spend so much time to keep āitā aligned.and you will have no mental model of your project.
5 Likes
Interessting statement. I usually have to force Claude not to use GDScript and write C++ code instead. At least the Claude models from Opus and Sonet 4.6 appear to know Godot and GDScripting quite well.
I can understand that many people trying to use local LLMs get frustrated fast. I never got any usefull results from that either.
P.s. maybe I understood the quote above wrong. Making up for the programmer not knowing GDScript and Godot ? That would be something else.
1 Like
Yep, there itās about vibe coding, without knowing the basics and without validating the code, just prompting.
But itās the same for experienced devs, at some level it will get chaotic. Especially in game development with all the dependencies, the LLM will ignore your already developed functions and will duplicate code and variables, Itās too much information and the result will be context rot.
Do you read every line of code? At some point you will just accept the changes without reading the code. At that point you will prompt the LLM for every micro change instead of developing coherent systems.
1 Like
By what means?
What do you mean by what means ? How I want to persue the sim project ? or If I have the skill to do so ?
Other human devs?
1 Like
Who will continue to generate the code once āaiā canāt do it any more?
1 Like
That would be my task mainly. But since I will make it open source I guess anyone could clone and do what they want with it.
ā¦which will deteriorate the quality of your tool (Claude and friends or whatever), by feeding it back its own output.
No human will come and try to enhance what you put up on Github, thatās no fun, either they try to do what you did themselves (because they are rich enough), or look for a job fixing hallucinated production code that would still give them a chance at an actual income.
2 Likes
Brace yourself, because youāll suddenly have to start doing thinking on your own. And not just any thinking, itāll be unpleasant kind of thinking about heaps of spaghetti slop code your agents left you with.
Oh, and Iām sure microsoft already notified you that come June, github copilot will start billing per token usage. This could easily 10x those $39.
5 Likes