😸 New Siri... powered by Google?!

Welcome, humans.

It’s that time of the year again: Coca-Cola’s released another (divisive) AI-generated holiday ad! In case you missed it, this is building on a trend from last year…

Now compare this year’s, which features lots of bright-eyed (and bushy-tailed!) close-ups of adorable animals, to the following:

1. The 2024 version (and this alt version)
2. The 2020 version (with actual humans)
3. The classic version

Without question, this year’s is leaps and bounds better looking than 2024. But is it better than the original??

We can compare them across three aspects:

1. Image quality vs intent: does it look how the filmmakers intended?
2. Vibes vs intent: does it make you FEEL the way the filmmakers want?
3. End result: does it accomplish what Coca Cola wanted? (Buy Coke!)

Obviously, you can’t beat the magic of the O.G., with that warm, nostalgic, 90s film grain feel. And even though the 2025 one looks better than 2024, videographers and editors will have lots to quibble over (weird cuts, this or that detail being off, etc).

But does it capture the same sense of magic? You could argue the creators are waving away quality concerns by cutting to adorable close-ups of tiny of animals, which does help a lot. But by the end, do you feel that same cozy, inspired feeling you do watching the 1995 one (associating cozy holidays with a cold glass bottle coke in the process)?

Last but not least, did they achieve in their goal to make you to want a Coca-Cola? IDK about that, but CLEARLY they got the people talking!! WDYT? Vote and tell us below!

Which "Holidays Are Coming" ad did you like best, and why?

Pick your fave: the O.G, the 2024 experiment, the 2020 revamp, or the new 2025 upstart?

Here’s what happened in AI today:

1. Apple paid Google $1B annually to power Siri with Gemini until 2026.
2. OpenAI hit 1M business customers, the fastest-growing business platform ever.
3. NVIDIA's Jensen Huang warned China could win the AI race.
4. Xpeng demo'd its new robot with a “vision action model.”

Advertise in The Neuron here

Apple Just Paid Google $1B to Save Siri…

According to Bloomberg’s resident Apple whisperer, Mark Gurman, the company just finalized a deal to pay Google roughly $1B a year to power the new Siri using Google Gemini. That's right—the company that built its empire on seamless integration and "it just works" is essentially renting Google's brain to fix its broken voice assistant.

So, y’all gonna rename “‘Apple’ Intelligence” to “Google Intelligence” or what?

Here's what's going down:

• Specifically, Apple’s going to use a “custom” 1.2 trillion parameter Gemini system as an “interim” solution until they get their own 1 trillion parameter model up and running (reminder that parameters = the model's internal settings that store its learned knowledge and skills),
• This model will handle Siri's “summarizer” and “planner” functions (basically the complex stuff current Siri struggles with).
• Apple will run Gemini on its own Private Cloud Compute servers, keeping user data completely walled off from Google (so Google gets a billi, but not your data).
• The new assistant (codename “Linwood”) launches next spring in iOS 26.4.

Apple doesn’t expect this to last forever. They're hoping to replace Gemini with their own 1 trilli model by 2026, and placed hey put some of their rockstar talent on the project (software chief Craig Federigh and Vision Pro creator Mike Rockwell).

So how’d Google pull this off? Apple tested it, ChatGPT, and Claude before finally choosing Gemini. The article doesn’t say this, but our guess is Gemini was best at following instructions and maintaining long context consistency.

Why Apple's doing this: Because Siri's been embarrassingly bad for years, and Apple's own AI team can't keep up. The company lost somewhere between 4 to 7 major researchers from Meta and competitors over the past few months, and event got sued for promising Apple Intelligence before it was ready.

On the flip side, the deal highlights Google's deep, full-stack AI advantage. The very Gemini model Apple is licensing is trained and served on Google's own custom Tensor Processing Units (TPUs). Coincidentally, Google just launched its 7th-generation
“Ironwood” TPU this AM, which is 4x faster and purpose-built for the “age of inference”; exactly the kind of massive-scale, responsive workload that powering Siri requires.

There’s just one problem… This deal won't work in China because, well, Google's banned there. Instead, Apple's pursuing partnerships with Alibaba and Baidu for a localized version, complete with government-approved content filtering, as one does.

Why this matters: Apple just paid $1 billion to basically rent someone else's technology for its flagship AI feature. That's either a pragmatic short-term fix or a sign that even Apple can't build world-class AI fast enough. Probably both.

The real question: Will Apple's in-house model actually match Gemini by the end of 2026, or will this “interim" arrangement become permanent? IDK about you, but if I had a $1B a year contract of any kind, I’d try to make that as permanent as possible…

FROM OUR PARTNERS

Why Sweden chose Sana for 2.3M people

Sweden needed agentic AI that could work for 2.3 million people—students, educators, and public servants. They chose Sana Agents.

Sana’s powerful AI agents operate alongside teams to get work done. They can automate workflows, analyze your data, create presentations, and take actions across 100+ tools—all grounded in your company's knowledge.

The results?

• 50% time savings in R&D analysis.
• 66% fewer hours on compliance audits.
• 95% faster responses to product questions.
• 2x more customer support issues resolved.

Your team deserves the same AI transforming an entire country.

See how it works

Prompt Tip of the Day

Here’s a downright stupid prompt trick that actually works stupidly well: when you ask the AI to help you write a prompt, and after you get your first result, ask it to try again but “make it at least 10x better.”

Here’s the exact language (and example outputs) shared by Reddit user Prestigious-Cost322:

Can you make this existing prompt at least 10x better right now? Do you have the capability to do it? Is there any way that it can be improved 10x?

A simple trick that can be applied in most situations for an instant increase in quality!

Treats to Try

*Asterisk = from our partners.

1. *Join Outskill's LIVE 2 day AI Mastermind: 16 hours of intense training on AI tools, automations & building agents to help you work smarter, earn more, and reclaim your time.
2. NotebookLM from Google now creates video summaries in anime and kawaii styles, lets you view your customization prompts for all outputs (audio, video, flashcards, quizzes), and is adding Google Sheets and image support soon.
3. Windsurf Codemaps generates AI-annotated maps of your codebase that show exactly where features live—ask “how does auth work?” and get grouped, nested sections with direct links to relevant code lines, plus visual diagrams.
4. Yansu generates design specs and simulates real scenarios before writing code, then traces every line back to requirements and team knowledge.
5. llama.cpp now has a ChatGPT-like interface that runs entirely on your laptop without internet; you can chat with 150,000+ different AI models, drop in PDFs and images, and branch conversations, all while keeping your data completely private and local (here’s how to set it up).
6. Maya1 generates expressive speech with 20+ emotions (laugh, cry, whisper, angry); just describe a “40-year-old warm British villain” and it creates that voice, running on a single GPU (AI chip) with real-time streaming.
7. WikiBoard turns Wikipedia browsing into a visual canvas where you can see all your articles laid out spatially, draw connections between them, and add notes (can we get a version of this for the entire web??).

Around the Horn

• Unitree demo’d a feature to turn their signature G1 robot into “embodied avatars” that mimic your movements in real time so you can have real time robo-fights like Real Steel!
• OpenAI hit 1M business customers (fastest-growing business platform in history) with companies like Amgen, Morgan Stanley, T-Mobile, and Target using ChatGPT for Work (which also hit 7M+ seats, up 40% in 2 months);
• Perplexity says it created custom kernels that let trillion-parameter models like Kimi-K2 run on AWS EFA, which = the first solution to make these massive models viable on AWS infrastructure.
• NVIDIA CEO Jensen Huang warned “China is going to win the AI race” due to its lower energy costs and fewer regulations that give it inherent advantages (we’d add the fact that it has a thriving open-source ecosystem and emphasizes building strong industrial bases around industries to that list).

FROM OUR PARTNERS

Breaking: The Great Startup Debate: 996 vs. The 4-Day Work Week

Silicon Valley’s founders are split. Some swear by the 996 grind 12-hour days, 6 days a week to chase market dominance. Others claim the 4-day work week makes teams faster, more creative and saner.

On Nov 19, Rippling is bringing both sides to the (virtual) stage for a live, unfiltered debate moderated by Varun Rana. You decide who’s building the future of work.

🎥 Expect bold takes, live Q&A, and exclusive giveaways (Airbnb gift cards + “First Principles” hats). Promo terms apply*.

Register Now!

Thursday Trivia

One is real, and one is AI. Which is which? (vote below!)

A.

B.

Which is AI, and which is real?

Which is AI, and which is real? The answer is below, but place your vote to see how your guess everyone else (no cheating now!)

A Cat’s Commentary

Trivia answer: A is AI, and B is real, but TBH… we’re not 100% sure. A could ALSO be real… We’re more sure about B. As commenter TV-VCR-Repair said: “This looks like exactly something my black cat would do if I tried to put a shirt on him. Let alone a Dolphins shirt.” LOL Sorry Dolphins fans!