In partnership with:
Dear Readers,
Today, a single CEO's "no" to the Pentagon might redraw the boundaries of AI power forever. Dario Amodei rejected a $200 million defense contract rather than let Claude be used for mass surveillance or autonomous weapons, and what happens after the Friday deadline could set the rules for every AI company building for governments worldwide.
But that's far from the only seismic shift in today's issue:
Isomorphic Labs just unveiled a drug design engine that more than doubles AlphaFold 3's accuracy
Epoch AI published a fascinating (and unsettling) analysis suggesting AI software efficiency may be improving up to 10× per year
Google dropped Nano Banana 2 to put pro-grade image generation in front of hundreds of millions of users for free; and
Burger King is now using an OpenAI-powered headset to police whether employees say "please" and "thank you" - because apparently the dystopia comes with a side of fries.
Grab your coffee and dig in.
Superintelligence X GTC
Yesterday, NVIDIA released its latest quarterly results - and once again, they exceeded all analyst expectations.
With that in mind, I'm particularly curious how Jensen Huang will frame this momentum at this year's GTC keynote. Because GTC isn't just a product launch event. It's where NVIDIA defines the narrative for the next phase of AI.
Will we see deeper details on Vera Rubin and the post-Blackwell roadmap? More clarity on inference economics? Updates on NVL systems and sovereign AI factories? Or perhaps a broader vision for the next compute scaling frontier?
Historically, Jensen uses GTC to do three things:
Introduce the next architecture leap
Expand NVIDIA's vertical integration across the AI stack
Reframe what the industry should focus on next. After another blowout quarter, the stakes feel even higher.
I'll be covering the keynote live from the ground this year, my first time attending GTC in person.
If you want to follow the keynote live from home, you can register via the following link: [NVIDIA GTC 2026 Keynote]
All the best,
Kim Isenberg
🧬 IsoDDE Redefines AI Drug Discovery
Isomorphic Labs unveiled IsoDDE, an advanced AI drug design engine that more than doubles AlphaFold 3’s accuracy on the hardest protein–ligand benchmarks. It also beats physics-based methods like FEP in binding affinity prediction at a fraction of the cost and time.
IsoDDE delivers 2.3x stronger high-fidelity antibody–antigen predictions and can identify novel druggable pockets using only amino acid sequences. This signals a major step toward fully in-silico drug discovery at scale. (Note: Isomorphic Labs is an Alphabet-owned company closely linked to Google DeepMind, led by CEO Demis Hassabis).
🚀 Software progress reshapes AI timelines
Epoch AI argues that “AI software progress” mostly means cutting the training compute needed to reach the same capabilities. The best (though highly uncertain) evidence suggests efficiency could improve several-fold per year, with a personal best guess hovering around 10× per year (and a wide 2×–50× credible interval).
The twist: A lot of what we’ve been calling “algorithmic progress” may actually be better data curation, synthetic data, and scale-dependent innovations that only pay off when training larger models. This means past estimates might be measuring the wrong thing. This reframes the “AI intelligence explosion” debate: correcting for underestimated software progress makes runaway feedback loops more plausible, but recognizing compute bottlenecks pushes the other way. The overall verdict remains genuinely unclear.
🍔 Burger King’s Headset AI Policing Politeness: AI dystopia arrives
Burger King is rolling out an OpenAI-powered headset chatbot called “Patty.” The system helps employees with kitchen tasks and explicitly tracks whether they use “friendly” phrases like “welcome to Burger King,” “please,” and “thank you.”
Managers can query the system for “friendliness” performance. While BK frames this as a coaching tool (and is working on tone analysis), it highlights a new era of granular workplace surveillance. Patty is part of the broader BK Assistant platform that connects drive-thru audio, kitchen equipment, inventory, and a new cloud POS. If something breaks or runs out, the system automatically removes items from menus across kiosks, drive-thru boards, and digital ordering within 15 minutes.
BK says it is cautious about full AI drive-thrus, currently testing them in fewer than 100 restaurants, but Patty is piloting in 500 locations, with the web/app platform targeting all US restaurants by the end of 2026.
The AI agent inflection point
Google's Fastest Image AI Yet
The Takeaway
👉 Nano Banana 2 (Gemini 3.1 Flash Image) combines Nano Banana Pro's advanced quality — 4K resolution, character consistency, precise text rendering — with the speed of Gemini Flash, making it Google's most capable and accessible image model to date.
👉 The model is now the default across Google's entire product suite including the Gemini app, Search, Flow, and Ads — replacing the original Nano Banana and putting pro-grade generation in front of hundreds of millions of users at zero additional cost.
👉 Built-in web search grounding means Nano Banana 2 can accurately render real-world subjects, locations, and data — turning it from a creative toy into a practical tool for infographics, marketing mockups, and localized content.
👉 Every image generated carries SynthID watermarks and C2PA Content Credentials, reinforcing Google's bet that provenance and transparency will become table stakes as AI-generated media floods the internet.
Google just dropped its most exciting image generation model yet - and it's fast enough to keep up with your creativity. Nano Banana 2 (officially Gemini 3.1 Flash Image) takes the studio-quality output of Nano Banana Pro and straps a rocket to it, delivering results at Gemini Flash speed. Think of it as Google handing everyone a pro-grade design studio that runs in seconds, not minutes.
The numbers are impressive: support for resolutions up to 4K, consistency across five characters and 14 objects in a single workflow, and the ability to render legible text directly into images - a notorious weak spot for AI generators. The model also taps into Gemini's real-world knowledge base and live web search to accurately depict specific subjects, making it a genuine tool for infographics, marketing assets, and data visualizations.
Nano Banana 2 is now rolling out as the default across the Gemini app, Google Search (AI Mode and Lens), Flow, Google Ads, and the developer API. Google AI Pro and Ultra subscribers still get access to the more specialized Nano Banana Pro model. Every generated image ships with a SynthID watermark and C2PA Content Credentials - a nod to Google's growing commitment to AI transparency.
Why it matters: Nano Banana 2 democratizes pro-level AI image generation by making it fast and free across Google's ecosystem. For creators, marketers, and developers, the gap between idea and polished visual just got dramatically smaller.
Sources:
🔗 https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/
In 2026, there's finally a better alternative to calendar links…
Scheduling meetings seems so simple, but it’s somehow weirdly broken... until now! I’ve been watching SkipUp recently, and it’s a total shift away from the clunky 'link era' that forces the recipient to do all the work. We’ve started using it because time zones wreck us, plus, the automatic follow-up is amazing for lost threads! Instead of the back-and-forth dance, you just CC [email protected] and let the AI handle the rest.
I think it's an impressive, white-glove experience that considers personal preferences, buffers, and time zones. One of the best parts is that it can connect to as many calendars as you like, to instantly check availability across your entire schedule. It provides the same professional touch as a high-end EA without the massive overhead, making it a perfect fit for lean teams and high-output operators. Seeing the clean, email-first output makes it clear that we don't have to choose between automation and etiquette anymore. If you're looking for real leverage over your time while maintaining a professional image, SkipUp is definitely the platform to watch.
If you're looking for real leverage over your time while maintaining a professional image, SkipUp is definitely the platform to watch.
Driven by investments in AI, hyperscaler capital expenditures have grown 70% per year since the release of GPT-4, nearing half a trillion dollars in total during 2025.
Anthropic Just Defied the Pentagon
The biggest standoff in AI history isn't about technology, it's about who gets to set the rules. Anthropic CEO Dario Amodei has rejected the Pentagon's "best and final offer," refusing to let Claude be used for mass domestic surveillance or fully autonomous weapons. The stakes are enormous: a $200 million contract, Claude's unique position as the only AI on classified military networks, and potentially the future of AI governance itself.
Defense Secretary Pete Hegseth set a Friday 5:01 PM deadline for Anthropic to accept "all lawful purposes" usage — or face contract termination, designation as a "supply chain risk," and possible invocation of the Defense Production Act. Amodei's response was sharp: the Pentagon's threats are "inherently contradictory" - one calls Anthropic a security risk, the other calls Claude essential to national security. Meanwhile, xAI has already signed a deal to bring Grok into classified systems with no restrictions, and Google and OpenAI are in accelerated talks to do the same.
This isn't just a contract dispute. It's the first real test of whether an AI company's voluntary safety commitments can survive when a superpower pushes back. Whatever happens after that Friday deadline will set a precedent for every AI company building for governments worldwide.
This standoff will define whether AI companies can maintain safety guardrails when governments demand unrestricted access. The outcome could reshape how every frontier AI lab navigates the tension between commercial defense contracts and self-imposed ethical limits.
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.