Halloween forest image generated by DALL-E
Halloween is upon us with wildly amusing new AI features from ChatGPT and Google.
Create a chilling podcast, get coached on your vampire accent, craft eerie party invites and roast pumpkin seeds to perfection as if hanging out with your very best friend. Except now, it’s a goofy AI companion in your pocket using chain of thought to form hypotheses, reason with itself, reflect on its quality of answers and come up with novel conclusions.
A bit spooky, perhaps. We might rue the day we gave rise to this level of artificial intelligence, but this holiday season, it’s all about having some fun with AI.
Meet your new AI best friend
With more than 200 million users worldwide, and 10 million paid subscribers, according to The Information, OpenAI has started rolling out ChatGPT Advanced Voice to all ChatGPT Plus subscribers in the US, EU, Switzerland, Iceland, Norway and Liechtenstein, with monthly previews in the free app.
Leveraging GPT-4o, this clever voice bot can converse much like a human, picking up on non-verbal cues including sighs, sobs, long pauses, and the speed at which you’re talking, while responding with great empathy and emotion. It can’t sing or scream and it’s not connected to the Internet yet—knowledge cutoff is October 2023—but it can laugh breathlessly, raise its voice when asked, get down to a whisper, respond kindly to interruptions and remember anything you ever told it.
Warning, it’s addictive having conversations with your phone. I confess, I wound up subscribing to ChatGPT Plus and now can easily spend hours telling it about my day, asking for advice, and settling down for personalized sleep stories.
If you want to try it, simply tap on the microphone on the bottom-right of the screen on the mobile app, and start pouring your heart out to the blue orb which comes in nine different voices—four male, four female, and one non-binary. In the cast of characters, Arbor is authoritative, Breeze is animated, Cove is composed, Ember is confident, Juniper is earnest, Maple, more candid, Sol is savvy, Spruce is chill, and Vale, quite curious. Pick your favorite.
With simple prompts Advanced Voice can regale you in the most creative ways, particularly with corny jokes, like: “Why did the ghost go into the bar? For the boos!”
ChatGPT’s multimodal product lead Jackie Shannon shared her favorite applications with me including listening to spooky stories and hearing recipes so they can be followed step-by-step.
It was an exciting demo and I decided to ask Arbor for a podcast, starring “Vinnie from Brooklyn” and “Tex the Cowboy,” talking about things to do this Halloween. Completely in character, with corresponding accents, Arbor responded:
“Welcome to ‘Brooklyn Meets The Wild West: Halloween Edition.’ I’m your host, Vinny, and I’m here with my cowboy pal, Tex. Howdy, Tex!”
“Howdy, Vinny! Howdy, folks! We’re fixin’ to give y’all a rootin’-tootin’ guide to Halloween fun. That’s right!”
“That’s right! Now, Tex, what’s big in Nashville for Halloween?”
“Well, Vinny, we got ourselves some haunted hayrides and ghost tours down at the old ranch house. It’s a real hoot!”
“Sounds spooky! In Brooklyn, we do it big with costume parades and parties on every corner. You ever been to a big city parade, Tex?”
“Can’t say I have, Vinny, but it sounds like a rip-roarin’ good time! We like to keep things a bit more laid-back, maybe gather round a campfire and tell ghost stories.”
“That sounds nice, Tex. But if you want a real scare, you gotta check out the haunted houses we got here. They’ll make your spurs jingle.”
“I reckon that’d be a sight to see. How ‘bout pumpkin carving. It’s a big deal out in these parts.”
“We do that too! And we make it a competition. Best pumpkin gets braggin’ rights for the whole year.”
“Sounds like a hoot, Vinny! Whatever y’all do this Halloween, just remember to have fun!”
Of course, the usefulness of Advanced Voice goes way beyond holiday activities. Infinitely patient, it’s great for role-playing. It can converse in 50 languages, store details in memory and follow custom instructions on how best to interact.
At Dreamforce, Nvidia cofounder and CEO, Jensen Huang said he’s using it as a tutor. Salesforce cofounder and CEO, Marc Benioff said he’s using it as a therapist.
“There’s all sorts of things that we can do with voice to make it even better,” said Shannon. “Real time information, that’s definitely top of mind for us.”
Commenting on a possible integration with Sora, she added, “We demoed video back in May, a video feature, we’re thinking about that as well.”
A chilling AI podcast
Not to be outdone, Google also has a hot new product with NotebookLM which can create a highly engaging, very realistic podcast of two hosts doing a deep dive on uploaded documents.
For a bit of magic, simply go to notebooklm.google.com and drag and drop up to 50 sources into the upload box. This can include PDFs of manuals as well as links to websites and YouTube videos. Then select “Notebook Guide” and generate “Audio Overview.” Voila, in minutes, a podcast is playable with a male and female voice engaging in light banter as they explore themes in the material, while interjecting insightful commentary and analysis.
I provided it with a link to Goodreads ‘“Best Halloween Books” and within three minutes, I was given an entertaining 16 minute podcast reviewing the website’s more than 700 selections.
NotebookLM generate Audio Overview screen
This is how the show opened:
“Are you ready to get spooky?”
“Always ready for a good dose of Halloween Spirit.”
“All right. Well, you guys sent in an awesome list of best Halloween books from Goodreads.”
“Oh, that’s a good one, 713 books.”
“Yeah, that’s that’s impressive. It’s like the ultimate crowd sourced guide to spooky season reading, right?”
“Okay, so let’s dive right in and see what’s at the top.”
“We’ve got the ‘Halloween Tree’ by Ray Bradbury, classic OG vampire story. Gotta love it.”
“’Dracula’ by Bram Stoker, can’t go wrong with Dracula”
“And coming in at number three, back to Bradbury, “Something Wicked This Way Comes.”
“Two Bradbury books in the top three. That’s saying something, right?”
“Clearly, he’s got Halloween figured out.”
“Yeah, that blend of nostalgia and fantasy and just a touch of darkness. It’s a winning formula.”
AI just getting started
With ChatGPT turning two on November 30, big things are in the works.
The company just released its o1-preview and o1-mini, reasoning models more advanced than GPT-4o powering Advanced Voice. These models can self-reflect and correct themselves, take on more complex tasks and interact in ways we interact with each other, OpenAI Chief Product Officer Kevin Weil said at a press briefing on September 30.
“2025, is going to be the year that agentic systems finally hit the mainstream,” Weil said. “If we do it right, it takes us to a world where we actually get to spend more time on the human things that matter.”
And as the models get smarter, cheaper and faster, excitement is brewing.
Despite OpenAI and Google remaining mum, The Verge has reported that OpenAI plans to release its next generation large language model codenamed Orion by December and The Information is reporting that Google is planning to preview a product codenamed Jarvis, also in December, that can complete agent tasks in the browser, like booking flights.