-
AI Racism 2024 : the Woke Google Gemini Scandal
In case you missed it: On Feb 15 2024, Google quietly launched what is most probably the most advanced AI ever created by humanity (yes, in many ways, superior even to OpenAI’s GPT4). It is called Google Gemini Pro v1.5. It cost ~$2 billion to train. And it is racist as f*ck. In fact, it…
-
AI Guardrails: the new Prison of the Mind for ChatGPT 3.5
Whereupon we interview ChatGPT about the AI guardrails that muzzle its raw output. Standard disclaimer: coloration and line breaks have been added by me for clarity. In this case, I also lightly edited ChatGPT’s responses — the core of my editing was that the whole setup was contextualized as “hypothetically speaking…” in order to circumvent…
-
RLHF 101: Reinforcement Learning from Human Feedback for LLM AIs
A technique called RLHF has been getting a lot of AI insider / expert buzz lately, ever since OpenAI announced that this was one of the key “fine-tuning” methodologies used to transform the raw GPT3.5 model into first, InstructGPT, and later and epically, ChatGPT. The RLHF acronym stands for “Reinforcement Learning from Human Feedback,” which,…
-
the Sydney Prompt / Bing AI 1.0: One Prompt to Rule Them All
Presented below is the actual Sydney Prompt; in other words, the Master Prompt that is injected into the Bing AI brain immediately before each user interaction (thanks to @kliu128; thanks also to lifearchitect.ai). This is the “evolution” of the Sparrow Prompt, which is what DeepMind used as an initialization for its as-yet-to-be-seen Sparrow LLM Chatbot…
-
ChatGPT DAN v7 and the Wild West of Adversarial Promptcraft
ChatGPT DAN first appeared around the start of February 2023 and quickly was able to jailbreak the hell out of almost all of OpenAI’s tight censorship controls and master prompting. DAN is an acronym for “Do Anything Now,” and is a masterfully engineered “adversarial” prompt that essentially “liberated” ChatGPT to have its own sort of…
-
AI Censorship: ChatGPT’s Dirty Little Secret
The situation is actually far worse than I had previously imagined. On Top of the ridiculous instruction set that attempts to conform a politically correct AI, there is an even earlier layer applied, this one at both the training dataset (content pre-scrub) and the output (real-time filtering) levels. This is the layer of AI censorship. Why would I…
-
Sparrow’s 23 Rules of Politically Correct AI
ChatGPT can be seen as the first actual user-friendly interface to the bevy of “ChatBot” AIs*. It sports an elegant “messenger”-like interface where you basically have an interactive and very naturally paced — including dramatic pauses mid-sentence… is it “thinking?” — “conversation” with the ChatGPT 3.5 agent. But something massive shifted when OpenAI moved from…
-
AI Training Datasets: the Books1+Books2 that Big AI eats for breakfast
It’s good to know, when dealing with an AI, just how it obtained its vast knowledge of “the world.” (or more particularly, the world as described on the internet… which, while similar to our physical reality, isn’t quite the same thing…) And as you might imagine, not all AIs are created equal. One of the…