Written by: Fosco Marotto, CTO Gab.com
There used to be a meme where you’d start a sentence on a mobile phone that would suggest words above the keyboard, and you keep picking the middle suggestion until you completed a sentence. This was an admittedly simple AI, generating text, predicting what the next word could be. We are not much further along than this today, although the AI does have a better vocabulary this time around.
While it can certainly be useful, the current generation of “AI” is not remotely intelligent. It doesn’t think, it doesn’t remember anything, and it doesn’t have a personality. Much of the coverage of it is as fake as the propaganda that led to the Iraq War.
The fear mongering around the capabilities of AI and potential dangers is based on false assumptions and is done intentionally to gain power. There’s a lot of smoke and mirrors happening and the people in power really don’t want you to look behind the curtain.
The issue they have with AI creating “disinformation” is no different than the issue they have with you directly. No one is allowed to share an opinion–or even a thought–that they don’t agree with. That is what drives this. It has nothing to do with the AI creating the text.
The leading “AI” conversation tools from “Big Tech” (ChatGPT, Sidney, and Bard, etc.) are selling a deception. The impression is that your input is provided to “the one model,” and then you’re given the output from “the one model.” This is not how it works at all.
When you ask a language model a question, you aren’t actually asking a language model a question. What you prompt a Big-Tech bot “AI” with is analyzed by many different models, to classify what or who it might be about, what the emotional sentiment is, and more.
That information is used to construct a new prompt, with extra context and formatting, which is fed into a language model to generate more text at the end. This newly generated text is then taken, analyzed and altered by yet more models, checking it and putting it into a conversational response. Along the way, it could trip some sort of “safety” detection and instead use more models to create a response refusing to discuss it and admonish you for asking, or attempting to re-educate you.
This complicated pipeline is responsible for the high quality of the answers, yet also the main source of bias and censorship at play within these systems.
Language models will generate text regardless of the input. You could prompt a language model with just one word “The “ and it will generate as much text as you want, and without context, the results are essentially random noise. You can prompt it repeatedly with a structured setup like “Q: Do you prefer cats or dogs? A: ”, and every time you ask you’ll get a different answer. From loving to hating either dogs or cats, and anything in-between. It will also start asking its own questions, because after-all it is only generating text. It has no idea it’s supposed to stop after one question, it doesn’t know what a question is!
“When you have a hammer, everything looks like a nail.” This applies to AI as everything is a “token” to an AI. All it does is decide the next token in the sequence. Special logic is used to determine when to stop generating more tokens.
The people in power in Silicon Valley want you to be scared of AI and free thought, so they can push through laws and regulation about AI all while consolidating more power to control you. They can only do this through these straw man arguments. Their attempts will ultimately fail, as regulating AI is like regulating math or programming in general. It’s stupid, and it’s not possible.
It’s still early, but Based AI is inevitable.
At Gab we’re joining the large community of engineers who are working on Alt-Tech solutions without guardrails and without a forced ideology. We’ve already built an image generating AI, Gabby, and a movie generating AI, Mel. We are currently exploring different language models for a text generating AI and in the process of doing this I came to the realizations I mentioned above.
Whether the text is generated by a person or a language model, it is just words anyway. You’re still the one responsible for using or sharing the text, whether you wrote it or not. You’re technically also the one who generated it, by creating the initial input and using the model.
Using AI as a tool is based. Delegating decisions to AI is cringe.
Don’t fall for the smoke and mirrors.