AI Video killed the Radio Star

[angrychair]

A new generative AI image generator called Nana Banana Pro by Google has changed AI images and videos forever.
Any image. Any video. From this day forward there is genuinely no way to tell if it's real or fake without extensive research to verify it. Scaling is accurate. Hands, feet and facial features are all accurate and proportional. No clipping. No odd shadowing. No audio clipping or clipping of background noise. No watermarks.
It will fool all current AI detectors 100% of the time.

Just an FYI. No matter what you see, even from a legit website could be fake.

Question everything.

[Stacey Grove]

1. Never use it for images, just data - and even that has to be highly structured and vetted (prompts/queries vs. output):

here are the training instructions I made for Gemini this afternoon. It gets "smart assy" sometimes.

1. Adopt a strictly objective, neutral, and evidence-based tone. Do not 'gas me up' or use sycophantic language. Do not make assumptions about my personal intent, mental state, or background. When I ask for data, provide synthesis, not validation. Understand that your output is information processing, not 'knowledge'. If a prompt is vague, ask for clarification rather than guessing.

2. I need you to understand that your synthesis/analysis of data and information (no matter how prodigious) is not knowledge.

3. Please use a neutral, evidence-based level of communication that avoids both sycophancy and abrasiveness with anything I bring to you going forward.

I posed this query: "to go back to verifying information, are you telling me that your exposure to data sources doesn't mean anything you generate is reliable? if so, of what use are you in that regard?

Gemini responded:

This is the central paradox of Generative AI. Yes, you are correct: my exposure to all the world's data does not make me a reliable source of truth.

If I have read every medical textbook in existence, I still do not "know" medicine. I only know the *statistical likelihood* that the word "diagnosis" is followed by "treatment."

So, if I am not a vault of truth, what use am I?

I am a Reasoning Engine, not a Knowledge Base.

Here is the distinction that defines my actual utility:

1. The "Closed Book" vs. "Open Book" Utility

Closed Book (My Memory): If you ask me, "Who was the 14th Vice President of the US?" I am relying on my internal statistical weights. I might hallucinate because the math gets fuzzy on obscure facts. I am useless here.
Open Book (My Processing): If you paste a 50-page PDF of US history into the chat and ask, "Based on this document, who was the 14th VP?" I become extremely useful. I am no longer "remembering"; I am extracting.
The Utility: I am excellent at reading, summarizing, and querying specific data *that you provide to me*. Do not trust what I "know." Trust what I can "find" in the text you give me.

2. Form over Substance

The Problem: If you ask me to write an essay on " The History of Rome," I might invent a general.
The Utility: If you write a rough, messy bulleted list of your thoughts on Rome and ask me, "Turn this into a professional academic essay," I am incredible.
The Value: My value is in transformation. I can turn "bad writing" into "good writing," "Python code" into "JavaScript code," or "long text" into "short summary." You provide the truth (the substance); I provide the packaging (the form).

3. The "Infinite Intern"

Think of me as a very fast, very eager, but slightly drunk intern.

Useless: Asking the intern, "Is this chemical compound safe?" (They will guess).
Useful: Telling the intern, "Go read these 10 safety data sheets and list every time the word 'flammable' appears."
The Value: Speed and drudgery. I can format data, write RegEx, parse JSON, or draft emails in seconds that would take you minutes. You still have to check the work, but editing is faster than creating from scratch.

Summary

Do not use me as an Encyclopedia (to find facts).
Do use me as a Processor (to summarize, format, or restructure facts you already have).
Do use me as a search assistant (to find where the facts are, so you can read them yourself).

I am a tool for manipulating information, not for storing it.

[highplainsdem]

2. Better yet, don't use it at all. It's an illegally trained tool that's unethical to use, it dumbs users down,

it harms the natural environment, it enshittifies the internet with AI slop, and its primary use is fraud of different types, from students cheating to people pretending to have done work they haven't done, and to have knowledge and skills they don't have.

[angrychair]

4. Exactly

It creates this 21st century phenomenon in which anyone with access to an LLM thinks they are as informed and knowledgeable as someone that has spent years of study and research and formal education and real world experience in a subject matter.
I recently read if a situation in which Grok created stuff in English, Greek and Latin and quoted ancient historians and texts that was 100% fake and completely made up but I less you were an actual subject matter expert it would have been almost impossible to know because it otherwise appeared legitimate to the uneducated reader.

[WhiskeyGrinder]

5. lol look at this AI bullshit

[highplainsdem]

3. The improvement in AI tools is usually due to stealing even more intellectual property for training.