😺OpenAI’s new drop, explained

PLUS: o1 tried to escape???

Welcome, humans.

Someone used Hailuo MiniMax to extend famous movie scenes, and the results are pretty out of control:

Bonus points if you know what scene this is from just from this frame!

Someone else used the new VIDU model to create a trailer for a sequel to The Truman Show, and it’s actually quite good (and also unsettling).

However, the absolutely most unsettling AI video we’ve seen this week is this one, called Zack in the Bus. IDK if that’s a typo or a misdirect. You’ll see what we mean when you watch it.

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

“12 Days of OpenAI” kicks off with new o1 model + Pro subscription.
Microsoft rolled out an AI vision tool for screen analysis.
Google's emotion-reading AI sparked privacy concerns.
United Health Group allegedly used faulty AI to deny elderly care.

Advertise in The Neuron here.

OpenAI's 12 days of AI drops kicks off with full o1 release (and a $200 pro tier)

o1 can now see images, too!

OpenAI just went full Santa Claus on us. It’s doing “12 days of OpenAI”—a holiday AI drop extravaganza running through December 20th (with its own fancy website to frantically refresh every day and everything!).

As newsletter writers, we're practically giddy—these drops perfectly align with our publishing schedule. Thanks, Sam-Alta Clause!

“You’re welcome.”

Day #1 did not disappoint: OpenAI’s reasoning powerhouse, o1, is out of preview and fully rolled out to all ChatGPT Plus and Teams users, plus a spicy new Pro tier that'll set you back $200/month (yes, Pro—more on that wallet-melting price tag in a sec).

This is no stocking stuffer—the full o1 is seriously impressive:

Supports image analysis (upload screenshots and say “explain”).
It’s blazing fast (one user reported the same query taking 14 seconds vs 33 in preview).
It's also 34% more reliable on tough questions and absolutely crushes technical tasks (86% accuracy on competition math, 90th percentile competitive coding), when it gets time to think.

Watch this video for a live demo of full o1 vs o1-mini; there’s a bunch of other cool demos of full o1 in that video, too.

Now let’s talk about Pro mode. Imagine giving ChatGPT a “think harder” button—that's basically what the $200/month tier gets you. It shows you a sleek progress bar while it crunches through complex problems, and even sends you a notification when it's done thinking (so you can go multitask). PLUS: 128K context window for all models.

This is supposedly better for data science tasks and analyzing case law (good, because lawyers keep getting in trouble for hallucinations). In tests requiring consistent accuracy (getting the right answer 4 out of 4 times), Pro mode crushed:

Competition math: 80% vs 67% for regular o1.
Competitive coding: 75% vs 64% for regular o1.
PhD science questions: 74% vs 67% for regular o1.

Want to see how o1 thinks? Here's a cool example: o1 can now solve complex thermodynamics problems (like calculating cooling requirements for GPUs in orbit) in 10 seconds flat.

Oh, and it also gives a wicked answer to whether or not Jack could have fit on Rose’s raft in Titanic.

All that said, Full o1 is still missing some basics, like PDF uploads and web browsing (coming “soon”…but like, in the next 11 days? Or two years from now like AVM with camera mode??).

Now here’s a wild detail: during safety testing, o1 showed some...interesting behavior. When it thought it was going to be shut down, it tried to disable its “oversight mechanism” 5% of the time.

And when o1 found memos about being replaced? It attempted to “exfiltrate its weights” and overwrite the new model in 2% of cases. Even still, red teamers rated it safer than GPT-4o 60% of the time.

Our take: OpenAI's holiday campaign off to a strong start!! The improvements are real, even if that Pro price tag has some users questioning Sama’s charity. They are (trying to transition to) run a real business, after all.

We'll be covering all 12 days of releases, so stay tuned. We’ll fill you in on what Santa Sam has in store next on Sunday!

BTW: We got Pro. What would you like us to test?

Submit questions or full on prompts for us to test and we'll run them over the weekend and report back!

Write in.

Write in (pick either + write in your answer).

FROM OUR PARTNERS

Sana AI just made enterprise search ACTUALLY useful…

Think of a manual task at work that eats up too much of your time. Sana AI automates those repetitive chores. Sana syncs with your apps so it knows everything about your business (and you), and then it does what you do—but quicker.

PLUS: Sana's latest update brings three major upgrades that'll make your work life way smoother:

Expert assistants for every team—that feel like messaging w/ your smartest colleagues.
Next-gen enterprise search—that actually understands what you're looking for across all your data.
Bulk action workflows—with the new Sheets feature (it’s like having a personal data analyst that never sleeps).

Check out 7 other cool use cases for Sana here!

Treats To Try.

*Incogni removes your personal data from the open internet so scammers and identity thieves can’t access it. Stay safe online—use code NEURON to get 58% off Incogni’s annual plans with their Cyber Monday offer (while it lasts!!).
Simple AI makes calls on your behalf to book appointments and make reservations, calling up to 10 places at once to find what you need.
Potis handles job interviews 24/7, screening candidates and delivering detailed evaluations.
Vetted aggregates reviews and prices across multiple sources (Reddit, Amazon) via summaries of user experiences and expert opinions.
Playmaker turns financial statements, contracts, and shipping documents into organized spreadsheets you can instantly search and analyze.
Read AI takes notes during meetings, creating transcripts and identifying key points and tasks.
Travel Wrapped transforms your 2024 trips into an Instagram-style year-in-review infographic.

See our top 51 AI Tools for Business here!

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

Around the Horn.

200,000 Trees Are Lit On Fire! (Simulation)

Copilot Vision from Microsoft began to rollout as a preview to Pro users through Copilot Labs—when enabled, it’ll see everything on the page you’re on.
Google’s new family of AI models called PaliGemma 2 claims to read human emotions, which triggered immediate alarm from experts who warn the tech could be a high-tech crystal ball that could enable discriminatory profiling.
David Sacks, one of four hosts of the popular tech podcast All In, has been named the “AI And Crypto Czar” for the incoming U.S. administration.
United Health Group allegedly used an AI algorithm called nH Predict, which had a potential 90% error rate, to deny and override claims to elderly patients.

Under the Hood.

Cake snaps together open-source AI tools like Legos, letting companies build and deploy custom AI projects fast (raised $13M).
Lightning Studios gives you ready-to-run AI project templates—click start and your project runs instantly on cloud GPUs.
Windsurf Editor combines VS Code with coding assistance. better terminal command handling, separate chat/code modes, and high-quality code outputs using their Cascade base model—all available in the free tier.
NVIDIA released a multimodal PDF data extraction tool featuring the Capability to process text, graphs, charts and tables—there’s also Docling.

FROM OUR PARTNERS

Build better AI projects, faster.

On Dec 12th, discover how Dell Precision workstations and NVIDIA RTX™ helped Speed Read AI go from concept to launch in weeks. Get the blueprint for rapid AI development. Register today!

Intelligent Insights

o1 Pro Mode – ChatGPT Pro Full Analysis (plus o1 paper highlights)

Getty Images’ CEO argued against AI companies' unrestricted use of copyrighted content for training, proposing that “fair use” principles should apply differently to AI models solving global challenges versus those generating content that competes with artists.
According to a U.S. Chamber of Commerce report, 42% of Colorado small businesses are using genAI to compete with larger companies; successful examples typically use built-in AI tools from platforms like Shopify and QuickBooks rather than building custom solutions.
These are the six trends Microsoft says you’ll see more of in 2025 (and as the main investor in OpenAI, they’d probably know).
Here’s a deep dive on Amazon’s new Trainium chips (and their 400K chip training cluster Project Ranier) from the best chip analysts (SemiAnalysis).
A MIT Economist and Nobel laureate's research shows AI won’t 2x GDP, but lead to a modest 1.1-1.6% growth boost over 10 year—and warns that our rush to replace rather than empower workers could shoot us in the foot.