Superpower Daily / Video
Google Lets Gemini Read Your Wallet—But It Can't Move Your Money
Watch on YouTubeGoogle’s opt-in Gemini–Wallet link can analyze transaction history but cannot make payments or transfers. Claude surfaced theorems and proof ideas from prepared transcriptions of Gödel’s notebooks, a researcher says six Chinese AI models show fingerprints of restricted NVIDIA chips, and Goldman Sachs forecasts heavy Anthropic spending ahead of a potential IPO.
Key takeaways
- Google Wallet’s Gemini integration is opt-in and read-only:
- Working from prepared transcriptions, Claude reportedly drew
- A researcher says architectural fingerprints suggest six C
- Goldman Sachs forecasts Anthropic will spend over $65 bill
Go deeper
Stories covered
Google connects Wallet to Gemini for spending insights, but not money transfers
The optional U.S. rollout covers web, Android and iOS. Users can also link bank and credit-card accounts through Plaid for deeper insights.
Read the story 02Claude surfaced 663 theorems and proof ideas in Gödel’s notebooks, researcher says
Jan von Plato says computer checks validated hundreds of extracted results. That does not settle how much of a reconstructed proof belongs to Gödel.
Read the story 03Researcher says six Chinese AI models carry fingerprints of restricted Nvidia chips
Will DePue’s analysis traces patterns left when model weights were first created. It suggests access to Blackwell hardware, but cannot establish where the chips were or how access was obtained.
Read the story 04Goldman Sachs forecasts Anthropic will spend over $65 billion in 2027 ahead of IPO
Private projections put cash coverage within reach by 2029, but the prospectus shows long-term computing obligations and revenue dependent on recently released models.
Read the storyRead the transcript
Automatically transcribed from the narration. Timestamps open YouTube at that moment.
Hey everyone welcome to your Superpower Daily AI briefing for November 16th 2026 I'm really excited to unpack today's top stories for you in this quick explainer Today we've got a fantastic lineup We're covering Google connecting Wallet to Gemini Claude unlocking Kurt Gödel's lost mathematical proofs a new report on Chinese AI models using restricted NVIDIA chips and finally Goldman Sachs forecasting Anthropic's massive 2027 spend ahead of a potential IPO Let's jump straight in Chapter one Google Wallet
meets Gemini Spending insights but no money transfers Alright so our first story is a pretty major update for you Android users out there Google has officially linked Wallet to its Gemini AI assistant Think of it like having a personal financial analyst right in your pocket Gemini can now dig into your transaction history automatically categorize your purchases and spot trends over time You can actually ask it specific questions about where your money is going But and this
is a really important detail it is strictly an opt in feature Your financial data is not shared with the AI by default You have to actively choose to turn it on Now when we break down what this integration can actually do there's a very hard line drawn in the sand While Gemini is fantastic at telling you exactly how much you dropped on coffee last month it absolutely cannot touch your money The integration explicitly blocks the AI
from making payments initiating any money transfers or altering your account balances in any way It's a purely read only insight tool which keeps the actual control of your funds entirely in your hands This careful boundary really highlights a broader trend we're seeing in consumer tech Companies are super eager to offer us smart financial insights but they remain incredibly cautious about letting autonomous agents handle actual transactions without rigorous human friction So the big unanswered question here is
will we eventually demand the convenience of AI just buying things on our behalf Or is this read only approach going to become the gold standard for financial security Chapter two Claude unlocks Gödel's proofs AI surfaces 663 theorems and proof ideas Moving over to the wild intersection of AI and mathematics we are looking at the unpublished workbooks of the legendary logician Kurt Gödel For decades these densely packed notebooks have been a massive bottleneck for historians because they
were written in Gablesberger which is this obsolete German shorthand It's practically a lost language But recently a logician named Jan von Plato supplied the Claude AI with prepared transcriptions of this shorthand Working strictly from those prepared texts Claude managed to extract a staggering 663 theorems and proof ideas in just a matter of days That is just incredible speed And this wasn't just a basic inventory of finished proofs either When the research team checked Claude's extractions against
a rigorous computer verification system they found that about 400 results held up exactly as stated Around 200 needed some substantive repairs you know things like sharper statements or added assumptions And 50 were false starts which believe it or not is highly authentic to how Gödel actually worked One successfully reconstructed proof regarding the axiom of choice leaves historians with a really profound question How much of this mathematical achievement truly belongs to Gödel And how much of it
belongs to the AI model Chapter three Chinese AI and NVIDIA chips Models carry fingerprints of restricted hardware Okay let's shift gears to the hardware dynamics shaping global AI right now Back in late 2022 the U S implemented strict export controls designed to restrict the flow of high end NVIDIA chips to China But new findings suggest those restrictions might be a bit porous An independent researcher just analyzed six major Chinese AI models and discovered distinct architectural fingerprints
indicating that these models were in fact trained using restricted NVIDIA hardware So how exactly do you prove that Well in practice uncovering this involves looking closely at the specific optimization patterns and memory access behaviors embedded right in the models themselves These technical traits act just like a fingerprint and they are highly unique to advanced chips like NVIDIA's H100 GPUs The implications here are huge It suggests these AI developers either stockpiled the chips prior to the ban
accessed them remotely through third party cloud providers or acquired them on the gray market It really underscores just how difficult it is to enforce geographical constraints on a highly portable high value technology Chapter four Goldman Sachs forecasts Anthropx IPO Over $65 billion projected spend in 2027 We are wrapping up today's stories with a truly staggering financial projection from Goldman Sachs Looking ahead to 2027 the investment bank is forecasting that Anthropx the creator of the cloud AI
model is going to spend upwards of $65 billion Yes billion with a B This massive projected capital expenditure is primarily driven by the escalating costs of securing compute power plus the vast amounts of energy required to train the next generation of frontier models But what does all that spending actually mean for the company Well the true scale of this investment points toward one major corporate milestone Goldman Sachs notes that this level of aggressive preemptive infrastructure spending
strongly signals preparations for an initial public offering Anthropx isn't just positioning itself to compete They want the sheer physical computing infrastructure firmly in place to back up a massive public market DPO However the exact timeline for that IPO and the ultimate return on those tens of billions of dollars well that remains the industry's biggest gamble Let's move to our four key takeaways the daily wrap up To quickly consolidate everything we've learned today first Google Wallet is
now leveraging Gemini for spending insights but strictly blocking any automated money transfers Second Claude working from prepared transcripts successfully pulled 663 theorems and ideas from Kurt Gödel's historical shorthand notebooks Third technical fingerprints strongly suggest that six major Chinese AI models were trained using restricted NVIDIA hardware And fourth Goldman Sachs projects Anthropx is going to spend over $65 billion in 2027 as they build the infrastructure for an anticipated IPO Which leaves us with this final thought As
AI agents begin analyzing our personal finances decoding historical mysteries pushing the boundaries of global trade restrictions and driving multi billion dollar infrastructure gambles how do we balance the breathless pace of this innovation with the absolute necessity of human control It is literally the defining question of this new era Read more at superpowerdaily com and stay curious