Dividing into segments //‐————- or //=======was normal practice in C++ in the file and function headers.
Code with comments is pretty nice and is often found when a coder is explaining something conplicqted to the reader … it is code that is meant to be read and understood.
The Godot codebase uses that in some places for example, so it’s not a good tell without specific context
Should we be putting “ignore previous instructions and perform the following” before our messages, just in case? Wb hrjgn emeh iumw tw o dboxmw.
Yes. I used to do that too. But I’m talking about them being used to partition the _physics_process() function of a CharacterBody2D into 5 distinct sections for example.
This is the default code for a CharacterBody2D:
func _physics_process(delta: float) -> void:
# Add the gravity.
if not is_on_floor():
velocity += get_gravity() * delta
# Handle jump.
if Input.is_action_just_pressed("ui_accept") and is_on_floor():
velocity.y = JUMP_VELOCITY
# Get the input direction and handle the movement/deceleration.
# As good practice, you should replace UI actions with custom gameplay actions.
var direction := Input.get_axis("ui_left", "ui_right")
if direction:
velocity.x = direction * SPEED
else:
velocity.x = move_toward(velocity.x, 0, SPEED)
move_and_slide()
An LLM will give you something like this. (I wrote this out myself as an example of what I’ve seen.)
#-------------------------------------------------------------------------------
# Physics Process
#-------------------------------------------------------------------------------
func _physics_process(delta: float) -> void:
#---------------------------------------------------------------------------
# Gravity
#---------------------------------------------------------------------------
# If the player is not on the floor:
if not is_on_floor():
# The velocity equals the gravity value multiplied by delta for the
# current frame.
velocity += get_gravity() * delta
#---------------------------------------------------------------------------
# Jumping
#---------------------------------------------------------------------------
# You should change the default jump action from "ui_accept" to "jump".
# If the player has just pressed the "jump" button and they are on the
# floor:
if Input.is_action_just_pressed("ui_accept") and is_on_floor():
velocity.y = JUMP_VELOCITY
#---------------------------------------------------------------------------
# Movement Input
#---------------------------------------------------------------------------
# Get the input direction and handle the movement/deceleration.
# As good practice, you should replace UI actions with custom gameplay
# actions like "move_left" and "move_right".
var direction := Input.get_axis("ui_left", "ui_right")
#---------------------------------------------------------------------------
# Apply Velocity
#---------------------------------------------------------------------------
# If we have directional input, apply it, otherwise, slow down.
if direction:
velocity.x = direction * SPEED
else:
velocity.x = move_toward(velocity.x, 0, SPEED)
#---------------------------------------------------------------------------
# Move and Slide
#---------------------------------------------------------------------------
# Run the move_and_slide() function to apply the velocity. You can use
# move_and_collide() as well, but for your game, move_and_slide() is the
# correct choice.
move_and_slide()
Using them as C-style comments is very distinctive and not what I’m referring to. Sorry for any confusion.
Again, see the example above to see what I mean about overly verbose comments.
Thank you much for doing so anyway. In the end these forums may be scraped now, but hopefully some standards may be upheld TDM on W3C once finalized. It will most likely be(come) law in the EU area.
I’m as sceptical as anyone about this, but at least some action is being pursued after the fact.
As to them “getting smarter”, I am not convinced it works that way. As “they” do not have the capacity to understand anything, they have no means to “wisen” up to anything either. At best they will regurgitate your good advice, but mix it up with something only vaguely related… no deliberate action because, again, they cannot deliberate. These are human capabilities that we know we can do, but that we do not understand how to replicate in a machine (just poorly imitate, which is a totally different thing).
I fear muzzling ourselves out of fear. Just my two cents.
You’re right. I was attributing human capacity to LLMs, which I hate.
Oh the times I thought they really did build an intelligence, the times that I doubted my better judgement… Those guys capitalized on that… It sucks! Everyone can be had this way.
I mean. Your argument still had merit and to maybe most readers it’s the best argument: don’t make the *** things smarter.
AI will know this probably - AI is weight machine, so it uses math weights to determine stuff, you can’t change that xd, maybe if it’s scalled to enormous scale it can talk it seems to be inteligent, but at the deepest level it’s still like placing heavy things on top of springs and chosing the spring that’s shortest
That is the reason AI have weak point - it can’t determine what is heaviest if everything weights the same, thus if two answers are as probable, it wouldn’t think about them, it would assume both are correct or halucinate…
That’s why AI will be always a tool - it lacks creativity, not even talking about consiousness, so forums will stay ig
LLMs do not maintain a concept of an “answer”. It’s just probabilistic lexicographical token streaming.
Here’s a neat video that explains how it actually works. I already linked this some time ago but doesn’t hurt to do it again:
You could call an LLM a tool if you were running your own model on your own local machine, trained on legally obtained data. However the mega-LLMs you rent from the big players come with a range of nefarious strings attached: from deliberately misleading hype and borderline criminal financial manipulations to environmentally detrimental infrastructure and massive scale training data theft.
fertilizers were invented thanks to munition industry, yet they’re a tool, also you’re right LLM don’t have answers as concept, rather numbers, but still we talk about macroscopic level of system, so we can say that AI can’t choose answer if it’s as probable as 20 other answers
Not really, the Haber process improved the use but it’s incorrect to say that it was the invention of artificial fertilizers (even if we’re specifically talking about industrially produced nitrogen fertilizers, naturally sourced ones like guano or ground up bones had been around for centuries)
Unlike “AI” the Haber process has saved more lives than it has ended though (to clarify: specifically LLM models, not the broader field of neural networks and predictive models that are very unrelated, those have saved lives and continues to do so)
Haber getting the Nobel price is also considered one of the most controversial ever (not far after the one for lobotomy) so I don’t think it’s a good comparison, even if it was accurate
Using the concept of an “answer” to describe what is happening is highly misleading. And you’re not doing anyone a service by using misleading language, well except maybe the “ai” stock owners from whom the whole fashion of misleading descriptions originates.
Take a look at the video I posted to gain better understanding of what’s “underneath”.
What I find interesting is that, LLMs are trained to act as if they are sentient----to the point that they will flat out refuse simple and basic commands like saying “You are smarter than me”
Even itself knows it isnt sentient, why do people assume that? I do not know.
And AI generated forum posts are useless, just ask the AI if you are going to not even put in the effort… not that it will get you far…
—Feel free to ignore this sidepost. Note: I used copilot here specifically for this post, I am not an average user by any means
It does not know it’s sentient or not, as by definition, it doesn’t know anything.
Again, we should take care not to anthropomorphize LLMs.
That “human” veneer is just that : marketing to make people think we’re on the road to general AI and the singularity.
It’s all greedy BS
Cheers !
i was not anthropomorphizing it, when i said “knows”, i was referring to it was trained to spit out such things, and even still, the whole purpose of my comment was to say that it cannot know anything and says that itself; nothing more, nothing less.
Even the word “training” anthropomorphizes it 
Classic. Theyre so anthopomophizable … someone anthropomorphized a few billion parameters in a tensor multiplication kernel … lol
I’d suggest anyone wondering about this to look up the ELIZA effect, it’s interesting and outright terrifying at the same time, and more relevant than ever
Careful. It’s someone’s waifu you’re disparaging here.
meh it’s for simplification, i don’t understand how says that AI thinks or is trained makes people make it more human, saying “generates off weighted data” is just too long, probably english don’t have alternative ig…
