The Neuron
Posts
😺 747 word mega prompt

😺 747 word mega prompt

PLUS: Sarah Silverman is MAD at OpenAI 🤬...

Noah Edelman
July 10, 2023

Welcome, humans.

Threads (Meta’s Twitter clone) raced past ChatGPT to become the fastest product to 100 million users.

We should’ve known: celebrity latte updates are more important than superintelligent chatbots. ☕️🤖

Here’s what’s going on in AI today:

An awesome prompt we found for ChatGPT’s Code Interpreter.
Sarah Silverman is suing OpenAI and Meta.
Google’s medical AI is being tested in hospitals.
Journalists are raging over AI-generated articles.

Sponsor The Neuron | New Here? Subscribe!

ChatGPT’s Code Interpreter Isn’t Just Good With Data 👇

We’re obsessed with ChatGPT’s Code Interpreter.

It cranks out data analyses at the level of an average data analyst in seconds (plus it’s also always behind a computer screen — get it?!). It’s so good that some suspect it’s powered by a model even more advanced than GPT-4.

(We highly recommend this comprehensive prompt for generating the best visuals).

But it’s not just helpful with data. Code Interpreter uses code to understand your files which means it can also tinker with them. This includes:

basic video editing (see example).
extracting text from images.
crafting entire games (walkthrough).
designing QR codes.

And the tech industry’s all time favorite: file type conversions. It does this all by running Python scripts.

But here’s the catch: If you’re an expert in design/data/CS, Code Interpreter’s features might feel limited. And, like ChatGPT, Code Interpreter might try to bullsh*t you when confused, (often by creating nonexistent data).

Here are a few expert prompting tricks to harness Code Interpreter from our guru Prof Ethan Mollick:

Feed it text files or plain text, not PDFs (too long).
Encourage it to “just try” if it says it can’t do something.
Engage in back-and-forth vs. lengthy prompts.

P.S. We’re gearing up to teach a course on how to automate your entire workflows using Code Interpreter and GPT-4. Sign up for our waitlist here!

Authors Are Suing OpenAI & Meta For Copyright Infringement 📚

Sarah Silverman, by Midjourney

Comics once worried about hustlers stealing their punchlines. Now it’s AI behemoths.

Sarah Silverman and two other authors are suing OpenAI and Meta for training ChatGPT and LLaMA on their (copyrighted) literary works.

Their line of defense:

LLaMA is trained on ThePile, which is trained on a shadow library called Bibliotik, which is known to be “flagrantly illegal”.
Oh, and the models are like wayyy too good at summarizing their books.

This isn’t a one-off.

Just last week, two different authors sued OpenAI. Artists are locking horns with Midjourney and Stability AI. Hollywood writers are protesting against AI.

Here’s our take: It’s glaringly obvious that these models are gulping down content without asking permission.

BUT: it’s kinda like illegally streaming films — everyone does it and no one ever gets in trouble. Expect more lawsuits and perhaps a few settlements, but defending content in the digital realm is becoming a losing battle.

Around the Horn 🦄

Google’s medical chatbot, Med-PaLM 2, is being tested in hospitals.
A deepfake video ad of financial journalist Martin Lewis shilling an investment circulated on Facebook.
Gizmodo is getting pushback from journalists for its AI-generated articles.
How the folks at Mithral Security tricked an AI (PoisonGPT) into spreading fake news.

Treats To Try 🍿

GPT-Migrate lets you migrate migrate your codebase from one language to another one.
Train your AI model once and deploy on any cloud with NVIDIA and Run:ai.
Pi, your personal AI chatbot, is now available on iOS.
Generate cookies for your website using Biscuits.ai.

We curated the top 27 tools you need for work in our “Top Tools For Business”!