😺 OpenAI's secret weapon

Welcome, humans.

We were scrolling Hugging Face yesterday, and found this trending dataset of “awesome ChatGPT prompts.” So you know it got our attention:

Want GPT to act like a lunatic? There's a prompt for that. Someone actually created a prompt that makes ChatGPT spout completely meaningless, arbitrary sentences.

And that's not even the strangest one—there are also prompts to make it act like a tea-taster (complete with fancy flavor profiles), a fallacy finder that calls out logical errors, and even a fancy title generator for when you need to jazz up your boring old project names.

Many are actually useful too. A particularly clever prompt turns GPT into a text-based Excel sheet—perfect for quick data crunching without ever opening up a spreadsheet.

Here’s what you need to know about AI today:

• GPT-5 “Orion” rumors point to December launch, OpenAI denies.
• Waymo raised $5.6B for expansion.
• Google released an endless AI video game.
• AI only needs 200-600 samples to copy an artist’s style.

Advertise in The Neuron here.

GPT-5, or Orion, or something else is coming out in December.

That’s right folks, the successor to GPT-4, whether it will actually be called GPT-5 or not, is rumored to be released in early December.

Let’s review the facts on the ground:

1. OpenAI’s next frontier model is codenamed Orion.
2. Orion is supposedly 100x more powerful than GPT-4, and is being trained on synthetic data from o1.
3. Orion will first be released to partner companies; for example, rumors are that Microsoft will get access to Orion as soon as November.

Here’s the thing though—OpenAI says it’s NOT “releasing a model code-named Orion this year.” Sam called the report “fake news.” But if this news is so fake, why did Sam post this cryptic tweet a little over a month ago?

Now here’s the REAL question: with the success of o1, do we really need GPT-5 right now? Don’t get us wrong, GPT-5 would be amazing to have.

I guess the real test will be, what can GPT-5 / Orion accomplish that current AI models can’t? By that we mean, can it:

• Reduce hallucinations to >0.01%?
• Maintain its memory for longer, and solve the lost in the middle problem?
• ACTUALLY reason?

If so, then it’ll definitely be worth it.

FROM OUR PARTNERS

This is the Biggest Breakthrough in Next-Gen Tech Since iPhone—but you only have until October 30th to invest.

Fun fact: Did you know Disney's princess IP alone has generated $46.4B in revenue!? Wild.

Here’s another fun fact: a next-gen tech company called Elf Labs has won 100+ trademark battles for characters like Cinderella, Little Mermaid, and Snow White—yes, really—and has advanced proprietary tech to bring them to life in your living room through next-gen AR (without clunky headsets).

With their unprecedented compression and patented tech, they’re creating hyper-realistic, AI powered 3D worlds with beloved, billion-dollar characters.

Now get this—Elf Labs has opened up an investment opportunity to the public. The team has already done $6B+ in licensing deals in their careers, and they're launching three new franchises in 2025.

Time is running out, though: With less than a week left to invest in Elf Labs, the opportunity to become a part of this game-changing movement ends on October 30.

Invest now before it's too late!

Treats To Try.

1. *Is generative AI actually a good investment for businesses in 2024? Join AI experts from companies like Moderna and S&P Global on Nov. 14 at Section’s AI: ROI Conference—a virtual event for leaders looking to achieve tangible results with AI. Register for free today.
2. Perplexity now has a Mac desktop app that lets you search just by typing cmd + shift + P—this is the latest of many other features it launched this month (good thread recapping them here).
3. LangChain connects your data and tools to build smart assistants, and there’s now LangGraph to design custom workflows for how these assistants make decisions, and LangSmith to catch + fix their mistakes before they reach users.
4. PreCog picks the best model to answer your question between GPT-4o, Claude 3.5 Sonnet, and Gemini 1.5 (just launched, in beta testing right now).
5. Overlap turns your hour-long videos into 2-minute viral clips that capture the key moments (might still require some manual edits tho)—try it here.
6. Vidify creates a video out of product stills—very new, and fair warning, even in the promo video we can already see some video generation errors, so prepare to prompt a lot to get what you want.
7. Granola is a meeting notes app that’s so popular with VCs they decided to give it $20M (Mac only atm).
8. AIorHuman is a new guessing game where you chat with a partner and decide if they're humans or AI (and compete w/ others on a global leaderboard).

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

MIT Technology Review created this interactive AI Hype Index that ranks recent AI events from doom to utopia, and hype to reality.

• Waymo raised another $5.6B from Alphabet, a16z, Fidelity, and T. Row Price and will use the funds to expand into new cities and develop its AI capabilities for “business applications.” Read more about Waymo here.
• Perplexity defended its AI search practices, arguing that public facts shouldn't be paywalled while remaining open to collaborating with Dow Jones publishers.
• Google created Unbounded, which lets you play an endless character life simulation game where every interaction, environment, and storyline is uniquely generated for your gameplay experience (examples here).
• A new paper found AI only needs 200-600 samples of an artist or person in their training data to recreate their recognizable features.

FROM OUR PARTNERS

Fully Automated Email Outreach With AI Agent Frank

Agent Frank was designed to fully take care of prospecting, emailing and booking meetings for you, so you can focus on closing deals!

With Agent Frank you won’t need 20+ different tools anymore - just set up your Agent in 4 steps and get started.

Learn More

Sunday Special.

There were actually quite a few models dropped this week that we didn’t cover, so we thought it would be nice to do a round-up of them:

1. Janus 1.3B from Deepseek handles image and text tasks better by separating the image understanding and creation processes under one system (code).
2. There’s also DeepSeek v2.5, which combines both strong coding and general writing abilities into a single open source model; allegedly, it’s as good as GPT-4 and as cheap as GPT 4-mini—you can chat with it here.
3. Aya Expanse from Cohere translates your content into 23 languages while matching each language's natural tone and style.
4. OmniParser from Microsoft helps your agents better understand and interact with user interfaces by converting screenshots into detailed descriptions of what can be clicked and what each button does (code, read more).
5. Llama-3.1-Nemotron improves Meta's language model with NVIDIA's custom training to deliver more helpful and factually accurate responses.

A Cat's Commentary.