Mar 25
Today we continue our #eli5AI series by explaining AI "slop" vs "AI art"
What's the difference? Isn't all Ai output the same? In summary: No.
architectsecurity.org/2026/0…
(ELi5ai= explain it like i am 5: AI)
ALT AI output vs AI "slop" vs AI art: what's the difference?
1
126
24 Nov 2025
Think about AI like this: You’re watching a movie, you’re perceiving it in realtime. You’re understanding start to finish
AI doesn’t experience it that way, LLM sees an entire 2 hour move frontwards, backwards, sidewards in the time it takes us to read the title
#eli5ai
167
24 Nov 2025
How do you tell a fake image from a real one?
Often, just look at the fingers
Consider this generated image for a battery organizer case
#eli5ai
1
221
24 Nov 2025
What’s the difference between an AI and spellcheck?.
This sounds like a joke setup but the answer is “nothing”.
There is no difference. Every spellcheck is just an LLM.
AI ingests vast documents. It has learned (via probability) most people say “a blue truck” not “a truck blue”
#eli5ai
Glossary: LLM means “large language model”
248
24 Nov 2025
Every instruction you give trying to micromanage AI prompts using long, detailed instructions forces it to make “the least bad” choice
If I asked for “cute penguins” followed by “they’re wearing sweaters, eating charcuterie, they have an iceberg for a coffee table”. These are perfect instructions for a computer program or a human designer.
AI doesn’t really work like that. (I mean, technically, it does, but you’re gonna get perfect “cute penguins” without all the rest. And you can try again in a few nanoseconds if you don’t like the penguins)
#eli5ai
24 Nov 2025
Stop thinking of AI prompts like you’re a programmer, and start thinking of AI prompts like it’s the skilled professional you’re paying to do your work. You’re the CEO.
A lot of AI problems and complaints for users stem from giving it too many constraints
We think a long, complex prompt is better than a short, simple prompt, but the opposite is true!
Asking an AI for “a working piece of code to do task X” is infinitely better than asking for “a piece of code in python, it uses library A and B, it’s secure and protects privacy, it’s written in English, uses interCaPpED variable names, and make it look like I wrote it” will produce the *worst* possible outcome
This is not C programming, folks. We do not need to give explicit instructions or define memory space
Every variable you introduce is not helping. It’s hurting.
The LLM needs to compare its trillions of data points against your dozen restrictions and instructions while its computing what word comes next in each of the words it generates
TLDR: Ask AI for something but don’t micromanage for the best results. You can always refine it later.
#eli5ai
269
24 Nov 2025
Stop thinking of AI prompts like you’re a programmer, and start thinking of AI prompts like it’s the skilled professional you’re paying to do your work. You’re the CEO.
A lot of AI problems and complaints for users stem from giving it too many constraints
We think a long, complex prompt is better than a short, simple prompt, but the opposite is true!
Asking an AI for “a working piece of code to do task X” is infinitely better than asking for “a piece of code in python, it uses library A and B, it’s secure and protects privacy, it’s written in English, uses interCaPpED variable names, and make it look like I wrote it” will produce the *worst* possible outcome
This is not C programming, folks. We do not need to give explicit instructions or define memory space
Every variable you introduce is not helping. It’s hurting.
The LLM needs to compare its trillions of data points against your dozen restrictions and instructions while its computing what word comes next in each of the words it generates
TLDR: Ask AI for something but don’t micromanage for the best results. You can always refine it later.
#eli5ai
1
442
23 Nov 2025
AI is just trying to metaphorically win family feud / your email draft by picking the most common, likely answer. It gets a sort of digital endorphin rush called a “reward”. A “good boy!” For the algorithm when it does good
Like a puppy, it learns from this (it’s called RLHF, tldr human feedback)
#eli5ai
149
23 Nov 2025
Ever watched Family Feud? Wanna understand AI in 10 seconds
TV producers polled 100 ppl (or whatev) and contestants need to guess the most likely answer
AI has polled trillions of docs, sites, videos and is suggesting the answer
It isn’t a whole lot more complex tbh
#eli5ai
1
4
528
23 Nov 2025
A coin toss has a 50% chance of heads or tails. The computer is effectively doing billions of coin flips to determine the next word in a sequence #eli5ai
133
23 Nov 2025
So, maybe it’s a hot day, you have location services turned on. AI is doing millions of instantaneous correlations of what the dog might be
It isn’t watching the dog (probably), nor taking its core temperature or even its temperament. It doesn’t know the name of your dog
AI just resds the 1s and 0s of “the dog is”, looks at billions of possibilities for the next word, and suggests something.
This is autocorrect, this is AI
#eli5ai
23 Nov 2025
Let’s look at AI and LLM prediction models
someone says to you “the dog is” you think to yourself “what, is the dog hungry? Soft? Old? Tired?”
AI is doing the same thing. It’s thinking “we are drafting an email. What is common messaging in email? What is the dog likely to be?”
This is literally all AI is today. A model of prediction based on how probable a set of words is
#eli5ai
538
23 Nov 2025
Let’s look at AI and LLM prediction models
someone says to you “the dog is” you think to yourself “what, is the dog hungry? Soft? Old? Tired?”
AI is doing the same thing. It’s thinking “we are drafting an email. What is common messaging in email? What is the dog likely to be?”
This is literally all AI is today. A model of prediction based on how probable a set of words is
#eli5ai
1
759
22 Nov 2025
Here’s a perfect example of how AI stumbles bc it’s not perfect
I’m trying to find a lyric from a song “why do I stumble” (the irony is massive here)
Search assumed I was trying to figure out a physical ailment. Further search just got weirder.
PS if you’ve been using a Shazam you’ve been using Ai
#eli5ai
205
22 Nov 2025
AI is really, truly that simple. #eli5ai
Yeah there’s a bunch of math and code but it’s essentially the same way human babies learn language!
The computer encounters “he ain’t” but uses billions of resources to correct that error
It isn’t smart per se, it’s just using vast resources to create reasonable output the end user won’t be disappointed by
#eli5ai
1
157
22 Nov 2025
An LLM is simply mathematical understanding of the frequency of words and word patterns
I failed basic algebra in high school, so I’m not a math nerd
But you don’t need math to understand LLM or AI
LLM/AI ingests data. It’s not even words, it’s 0s and 1s.
The output from a search or spellcheck is just prediction based on the model it was designed on
For example. If I start a sentence with “hello” it’s reasonable that I’m going to say the first name of the intended person next. That’s ENTIRELY how ai works. Prediction
AI is not guessing I’m talking to Bradley because it’s smarter then me
AI is simply a computer algorithm seeing I just spoke to Bradley, I’m replying to an email where Bradley is the recipient, and it’s assuming I’m going to want to type his name next
LLM is about context. Do people usually use “he ain’t” or “he isn’t”. Do we more often place descriptive words before nouns “a truck blue” or “a blue truck”.
This is AI.
#eli5ai
1
221
22 Nov 2025
Every time you check your spelling in @Microsoft word eg? You’re using an LLM (large language model)
LLM is the building block of the processing done by what people call AI
#eli5ai
1
164
22 Nov 2025
Training data of models like @google’s Gemini is essentially absorbing the entire public internet and a lot of digitized books and journals
Consider how historical records and content like Mark Twain’s “Huckleberry Finn” could create algorithmic bias of race and ethnicity. Huckleberry Finn is widely discussed, shared, published online, available free.
An algorithm doesn’t know about racism. It’s simply reflecting ourselves. #eli5ai
1
136
22 Nov 2025
A large language model (LLM/AI) is a reflection of the training library it receives. This creates “bias”. Let’s say men are portrayed throughout recorded human history as strong and masculine.
If the training data contains millions of human examples of historical, societal, or cultural bias (e.g., more examples of men in engineering jobs than women), the AI will learn and repeat that bias because, to the model, it is the most statistically probable pattern to follow.
#eli5ai
1
1
200