Youtube comments of (@henrikkniberg).
-
248
-
239
-
173
-
131
-
126
-
80
-
69
-
55
-
33
-
31
-
30
-
28
-
26
-
24
-
23
-
21
-
19
-
17
-
15
-
14
-
13
-
12
-
12
-
11
-
9
-
9
-
9
-
8
-
8
-
7
-
7
-
6
-
Hard to explain, but I'll try.
Let's say we're making a spline that maps continentalness to yoffset.
One point might say Continentalness 0.1 => Yoffset 0.5. That means the yoffset is fixed and independent of the other noises if continentalness 0.1.
Another point might say Continentalness 0.2 => (depends on Erosion).
And that in turn leads to separate spline that maps erosion to yoffset. It could contain points like Erosion -1 => Yoffset 0.2 and Erosion 1 => Yoffset 0.5.
And any of those erosion spline points might in turn refer to a third spline that maps PeaksAndValleys to yoffset. For example Erosion 0.8 => (depends on PeaksAndValleys).
Another way of looking at it:
Me: "What should yoffset be here?"
System: "What is your continentalness?"
Me: "Continentalness is 0.1 here".
System: "OK then yoffset is 0.5".
....
Me: "What about over here? What should yoffset be here?"
System: "What is your continentalness?"
Me: "Continentalness is 0.2 here".
System: "Oh, then it also depends on Erosion. What is your erosion?".
Me: "Erosion is 0.8 here".
System: "OK then yoffset is 0.3".
...
And these are splines, so if continentalness was somewhere between, such as 0.15, then yoffset would be roughly the average of the two examples above.
In the actual game, we've configured so if continentalness is low then the yoffset is also low, regardless of other noises. That gives us oceans. The ocean floor is only affected by continentalness, not the other noises (with the exception of super low values which lead to Mushroom Fields islands). While if continentalness is above 0, Erosion and PeaksAndValleys start having a big impact on terrain, creating plateaus and rivers and mountains.
6
-
6
-
6
-
6
-
5
-
5
-
5
-
5
-
5
-
5
-
4
-
4
-
4
-
4
-
4
-
4
-
4
-
4
-
4
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
3
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
I partially agree with you. Many once-upon-a-time ”essential” prompt engineering tricks (such as write ”think step by step”) are redundant now with the latest models, because they do that by default. Even things like ”feel free to ask me questions before answering” seems to be less important, with recent updates chatgpt for example will tend to ask questions anyway instead of jumping to conclusions. I think there are two types of prompt engineering. Not sure what to call them, but one type is more like prompting hacks, things that are important now but maybe not in the future. Similar to how you once needed to know google search ”hacks” to get good results in the past, but now you basically just type whatever you want and google tends to give decent results. But there is another type of prompt engineering which I think will stay essential in the future. That includes things like knowing when and how to iterate, remember to phrase your high level goal, and knowing when to include which types of context. And knowing that you can ask it to generate content with specific formats, etc. So in short, yes, some aspects of prompt engineering are going become redundant as the models improve, while other aspects are probably going to stay relevant even in the long term. But I’m just guessing here, like everyone else :)
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
2
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1
-
1