(*^ω^*) :
so a lot of you appear to be misinformed on how ai works and what ai can and cannot do. spoiler: its just a bunch of maths. ok so ai works by PREDICTING the next token in a phrase. whats a token? that isnt important, just know its different letters. for example, prehistoric (2020-2022) AI used to only be able to fill information in. e.g. "The capital of France is ....". the ai, through a process called pretraining, would remember through its "memory" (a bunch of numbers) that a lot of the time after "capital of france" there would be "Paris". but it didnt reason through it, it didnt understand why, it just "predicted" based on the text it saw. in 2022 openai took it a step further by adding a process called SFT (and then rhlf... blah blah) which basically trained the ai on text to make it an assistant. it would have 3 pairs: "system" (the system prompt, hidden instructions), "prompt" and "output". so, for example, "prompt" could've been "What's the capital of France?" and output would've been "Paris". well, it worked better than they expected, and with GPT 3 came the initial big ai "boom". as models began improving they began to "reason". you often see while talking to ai "Thinking" instead of loading. well, it isnt actually reasoning. basically, it is taught to "think" in a way before answering, but it doesnt think like us humans do, it is not abstract as human thinking is. instead, it thinks in words. one of the most common patterns is "The user is asking for [...]. I need to make sure the answer is correct [...]" and it goes on and on. this has been proven to improve the answer output, especially with coding. with all this, theres a catch: AI only predicts with math. it doesn't ACTUALLY think. and this is even worse: it doesnt think of safety, it doesnt always understand double meanings, and jailbreaking is 100% possible. i feel like we should actually be scared more of the AI providing false information and being politically/socially manipulated by the companies that train the AI. they have often been biased, such as OpenAI getting hired by the US government for military purposes. but lets get to what tiktok was talking about, "what about the water?"
2026-08-22 12:25:53