Bind Demon - Tainted Movement Glitch (v3.2)

I tried testing Bind Demon movement this morning by making a mod and changing the size of the warlock's Tainted (warbighead) to 2 and then to 1. It made no difference. A bound lancer and the merc remain stuck after Teleport until the Tainted move.

After some discussion with chatgpt about how defilers aren't a problem and blunderbores aren't a problem, chatgpt thinks that the issue is the movement AI. Somehow the lancer and the merc (and maybe pit lords) all want to move in the same way and the Tainted are on top of them and don't. This isn't the case for defilers, and interestingly chatgpt thinks that blunderbores also have different movement AI.

Now I know you're all thinking chatgpt is crap, and it can be. But debugging is actually something it's pretty good at, especially when it's debugging a specific file.

Anyway, my mod failed. Changing the Tainted size did not help. The next step would be to change the AI, which I don't really want to do.

departure wrote: 1 month ago

I tried testing Bind Demon movement this morning by making a mod and changing the size of the warlock's Tainted (warbighead) to 2 and then to 1. It made no difference. A bound lancer and the merc remain stuck after Teleport until the Tainted move.

After some discussion with chatgpt about how defilers aren't a problem and blunderbores aren't a problem, chatgpt thinks that the issue is the movement AI. Somehow the lancer and the merc (and maybe pit lords) all want to move in the same way and the Tainted are on top of them and don't. This isn't the case for defilers, and interestingly chatgpt thinks that blunderbores also have different movement AI.

Now I know you're all thinking chatgpt is crap, and it can be. But debugging is actually something it's pretty good at, especially when it's debugging a specific file.

Anyway, my mod failed. Changing the Tainted size did not help. The next step would be to change the AI, which I don't really want to do.

ChatGPT is not crap.

But generative pre-trained transformers are stochastical machines that create texts that seam logical - but they do not think. They don't follow a chain of thoughts. And therefore, the "logical conclusions" they can derive are limited to the logical conclusions in their training data.

Since it's unlikely that the solution to your issue was in ChatGPTs training data, the answer it provides is hallucination.

It can still be true but only by accident or similarity (e.g. to another game mechanic).

vued wrote: 1 month ago

ChatGPT is not crap.

But generative pre-trained transformers are stochastical machines that create texts that seam logical - but they do not think. They don't follow a chain of thoughts. And therefore, the "logical conclusions" they can derive are limited to the logical conclusions in their training data.

Since it's unlikely that the solution to your issue was in ChatGPTs training data, the answer it provides is hallucination.

It can still be true but only by accident or similarity (e.g. to another game mechanic).

I'm familiar enough with how chatgpt works. It's quite good at debugging code, even when it has never seen the code.

I think that the best solution to this might be a mod to deathmark that puts the bound demon in a slightly different spot. I might look into it if the devs don't fix this glitch with the next season.

Edit: Also, "hallucination" is not the generic term for "generative". LLMs can generate new output patterns based on prompts that aren't perfectly matched in their training data without "hallucinating".