Super Prompt: The Generative AI Podcast – Details, episodes & analysis

With great power comes great responsibility. How do Open AI, Anthropic, and Meta implement safety and ethics? As large language models (LLMs) get larger, the potential for using them for nefarious purposes looms larger as well. Anthropic uses Constitutional AI, while OpenAI uses a model spec, combined with RLHF (Reinforcement Learning from Human Feedback). Not to be confused with ROFL (Rolling On the Floor Laughing). Tune into this episode to learn how leading AI companies use their Spidey powers to maximize usefulness and harmlessness.

REFERENCE

OpenAI Model Spec

https://cdn.openai.com/spec/model-spec-2024-05-08.html#overview

Anthropic Constitutional AI

https://www.anthropic.com/news/claudes-constitution

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

So what are notable Open Source Large Language models?  In this episode, I cover Open Source models from Meta the parent company of Facebook, a French AI company called Mistral currently valued at $2B dollars, in addition to Microsoft and Apple. Not all Open Source models are equally open, so I’ll go into restrictions you’ll want to know before using one of these models for your company or startup. Please enjoy this episode.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

“Does chatGPT possess human-like intelligence?” It turns out there's a right answer, and that answer is “NO”! Does this definite answer seem out of character for chatGPT which usually goes overboard  with fair and balanced views?  It did to me. That's the rabbit hole I explore in this episode.  By probing around this  accidentally-encountered guardrail, we discover the kinds of ethical issues chatGPT's creators  are concerned about. And I wonder out loud we can't just be friends with AI, by adopting science fiction writer's Isaac Asimov's Robot Laws created 80 years ago. Please enjoy this episode.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

Does chatGPT have a sense of humor? What if after Microsoft's acquisition of Open AI, the Onion ran the headline, “Microsoft renames chatGPT to clippyChat”? Would chatGPT find this funny? TL;DR LLMs are better at analyzing humor than creating it. Please enjoy this episode. 

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

How do you extract prohibited information from ChatGPT? What are Grandma and DAN exploits? Why do they work? What can Large Language Model (LLM) companies do to protect themselves?  Grandma exploits or hacks are ways to trick chatGPT into giving you information that is in violation of company policy. For example, tricking chatGPT to give you confidential, dangerous, or inappropriate information. "Jailbreaking” is a slang  for removing the artificial limitations in iPhones to install apps not approved by Apple. Turns out, there are ways to jailbreak LLMs. The tech companies supplying LLM as a service want to provide a safe, and legally-compliant environment. How can this be done without hampering the flexibility and usefulness of creative prompting?

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

What are AI hallucinations, and are they a feature or a bug? We start with the Top 10 categories of AI Hallucinations and examples, then explore how chatGPT might hallucinate an answer to the question, "What is the central theme of Blade Runner?" We end with chatGPT debating with itself whether AI hallucinations are bad or good for humanity. Which side wins? Tune in to find out.

In these solo episodes, I provide more definition, explanation, and context than my regular conversational episodes with guests. The goal is to bring up to speed, those new to AI.

Format: Letters read aloud.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

Using the prompt, "Why isn't Superman's suit Kryptonite-proof?", we learn how Large Language Models are trained,  why "self-attention" and the "transformer" architecture (which is what the T in GPT stands for) makes GPT-3 so powerful, the process of "inference", and how chatGPT generates answers to nerdy Superhero questions. After this episode, you'll be able to impress your friends by using the previously-mentioned AI jargon in complete sentences.

In these solo episodes, I provide more definition, explanation, and context than my regular episodes with guests. The goal is to bring up to speed, those new to AI.

Format: Letters read aloud.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

"How do ChatGPT, GPT-3, and Large Language Models (LLMs) relate?" 

That is the question we explore this episode via

1. Nursery rhyme
2. A satirical Friend's episode w/ Chandler, Joey, Ross, and Monica
3. Fairy Tale 

We also examine the hierarchal order of: artificial intelligence, neural network, large language model,  GPT-3, chatGPT. And why I got the order  wrong initially. Hint: I reversed chatGPT and GPT-3.

In these solo episodes, I provide more definition, explanation, and context than my regular episodes. The goal is to bring those new to AI up to speed.

Format: Letters read aloud.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

In these solo episodes, I provide more definition, explanation, and context than my regular episodes. The idea is to help  those new to AI get more out of my conversations with guests.

Format: Letters read aloud. 

I start each solo episode with a question. In this one, I ask, "How would you describe ChatGPT in your own words? " I answered it for myself, then asked chatGPT how I did. Mayhem ensues.

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.

I speak with scientist entrepreneur, Arijit Ray. Arijit is a PHD candidate at Boston University. We speak about generative AI, why it’s so hard to get DALL-E to create  the exact pizza we envision, how one goes from scientist to entrepreneur, and his startup, which is training AI to predict social media responses and run marketing focus groups. Please enjoy my conversation with Arijit Ray. 

For more information, check out https://www.superprompt.fm There you can contact me and/or sign up for our newsletter.