⚙️ The math behind the AI bubble

Good morning. We’ve talked before about the mounting evidence that we are in the throes of a growing AI bubble akin to the dot-com bubble of the late ‘90s.

A venture capitalist at Sequoia recently published a simple equation that adds even more credence to that impression. We dive into it all below.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

AI for Good: Improving breast cancer screening results

Source: Unsplash

Integrating artificial intelligence into the breast cancer screening process improved performance while reducing radiologists’ workload, according to the results of a recently published large retrospective cohort study. 

The details: Researchers at the University of Copenhagen examined 60,751 and 58,246 women before and after the implementation of AI, respectively. 

  • After applying AI, the cancer detection rate increased, the false positive rate decreased, the recall rate decreased and radiologists’ workload was reduced by 33%. 

  • The study, though, had several limitations, including a lack of data on how regularly the radiologists involved agreed with initial AI screens. 

Other studies (from Sweden) have found that double readings by radiologists improved with AI support and that a reading by one radiologist and an AI system was better than a reading by two radiologists. 

Why it matters: The reduction in recall rates resulting from the implementation of AI “likely led to a considerable reduction in the downstream clinical workload,” according to the authors. At the same time, with the addition of AI screening, “substantially more breast cancers” were diagnosed. 

AI isn’t magic. It’s math.

Brilliant lets you play with the core concepts behind things like LLMs and neural networks, helping you unlock your full potential in the world of AI. 

Thousands of bite-size, hands-on lessons in logic, math, data science, technology and more help you level up in minutes a day. Each one is packed with visual, interactive problem-solving that make even complex topics understandable. 

Join 10 million other proactive learners around the world and start your 30-day free trial today. Plus, readers of The Deep View can unlock a special discount on a premium annual subscription with this link

Proton launches a privacy-focused Google Docs alternative

Source: Proton

Proton — the tech company building an ever-lengthening list of cloud products with built-in data security — on Wednesday launched Proton Docs in a direct challenge to Google. 

It looks and operates much the same as a Google Doc: white background, live collaborative editing, text options, multimedia capabilities, etc. It’s just built with security in mind. 

  • Proton said that every file, photo, document, keystroke and cursor movement is end-to-end encrypted in real time. 

  • Proton said even the company itself doesn’t have access to user files. The company additionally stores its data in Switzerland where it is protected by strict privacy laws.

Proton said that Docs provides a way for users to get out from under Big Tech surveillance, which might be used to train AI products without user consent. 

Zoom Out: Google has said that it does not train its AI models on user data in Drive. Google’s privacy policy further states that it does not use Drive content for advertising purposes. 

Google Drive is also protected by a strong layer of encryption — the risk, though, is that Google retains encryption keys for all its files, meaning a successful hack could expose data stored in Drive. 

  • The launch comes as users are beginning to look more intensely for alternatives to the Big Tech platforms that have gained the habitual use of billions over the past 20 years. Changes to Meta’s privacy policy have sent millions of artists to Cara, a secure alternative; after updates to its privacy policy, Adobe users began to call for alternatives as well. 

“Far too often, online productivity suites require you to accept surveillance as a condition of use,” Proton said. “Docs is a milestone in our journey to building a better internet where privacy is the default … so you can break away from Big Tech services that exploit your data.”

  • Share 2 Days with the Brightest Minds in AI including scientists and engineers from OpenAI, Meta, Deepmind, and many more.

    • This week only -- Register using discount code: deepview24 for a $350 discount!*

  • Researchers say crypto hacking thefts doubled to $1.4 billion in 2024 so far (Reuters).

  • Fear and longing in Sun Valley (The Information).

  • Cohere CEO says GenAI will bring more profit to companies (CNBC).

  • Kenyan protestors are using AI in their anti-government fight (Semafor).

  • Need a website? Generate one with AI in seconds using Mixo.

    • For this week only, get 20% off all website costs using offer code deepview20 at checkout.*

China has filed more than 5x the number of AI inventions as the U.S. 

Source: Tencent

A recent report published by the World Intellectual Property Organization (WIPO) — the ‘Patent Landscape Report on Generative AI’ — found that China is leading the world in AI patents and inventions. And it isn’t really close. 

The details: The report found that the introduction of ChatGPT likely set off a wave of interest in generative AI. 

  • Between 2014 and 2023, 54,000 generative AI inventions were filed and 75,000+ research papers on genAI were published. A quarter of these inventions were filed in 2023 alone. 

  • Still, patents for genAI represent only 6% of total AI patents — but its share of that total has been steadily increasing since 2017. 

The largest shareat 38,000 patent families between 2014 and 2023 — of genAI inventions were produced in China. “Since 2017, China has published more patents in this field each year than all other countries combined.”

  • The U.S. comes in at a not-so-close second place, with 6,300 patent families filed during the same time period. 

  • Tencent is the top genAI patent owner in the world, followed by Ping An Insurance Group, Baidu and IBM. 

WIPO said that this is just the beginning: “We can expect a wave of related patents very soon, especially as the success of ChatGPT has driven innovation in a wide range of applications. We can only guess at what is to come.”

Effortlessly Schedule Developer Interviews with PestoAI

Imagine effortlessly adding top-tier developers to your interview calendar.

PestoAI can source and interview developers from around the globe, outperforming the average recruiter. You only pay when you succeed.

Using PestoAI feels like magic. Here's what happens behind the scenes:

  • Targeted Sourcing: Our AI curates candidates based on your job description, using a model trained by local market recruiters.

  • Company Pitch: We pitch your company to the best-fit developers.

  • Technical Screening: Candidates undergo a thorough technical interview using our proprietary technology.

  • Seamless Scheduling: We schedule the interviews directly on your calendar.

For Deep View readers, we are offering an exclusive 50% OFF if you sign up today!

The math behind the AI bubble 

Source: Created with AI by the Deep View

The basic concept that supports the idea that we are living through an AI bubble is simple: the revenue just isn’t there and the expense is enormous. 

  • Microsoft’s Azure cloud, for instance, grew some 26% in 2023, with only two percentage points deriving from AI services (Microsoft didn’t reveal the actual numbers). This came several months after Microsoft execs said they soon expect to see $10 billion in annual AI revenue. 

  • OpenAI — the leader in this race — recently topped $3.4 billion in annualized revenue, according to The Information. It’s unclear what kind of costs OpenAI incurs on a daily basis to train and run its models (some of which are free); The Information reported last year that in 2022, OpenAI’s losses reached $540 million. 

A single Nvidia H100 chip costs around $30,000. 

Here’s the calculation: Last year, Sequoia’s David Cahn explored AI’s ‘$200 billion question.’ 

The basic math is Nvidia’s datacenter run-rate revenue forecast multiplied by 2x, and the resulting number multiplied by 2x again. 

  • “For every $1 spent on a GPU, roughly $1 needs to be spent on energy costs to run the GPU in a data center. So if Nvidia sells $50B in run-rate GPU revenue by the end of the year, that implies approximately $100B in data center expenditures.”

  • “The end user of the GPU — for example, Starbucks, X, Tesla, Github Copilot or a new startup — needs to earn a margin too. Let’s assume they need to earn a 50% margin. This implies that for each year of current GPU CapEx, $200B of lifetime revenue would need to be generated by these GPUs to pay back the upfront capital investment. This does not include any margin for the cloud vendors—for them to earn a positive return, the total revenue requirement would be even higher.”

He said at the time that “even if you assume extremely generous gains from AI, there’s a $125B+ hole that needs to be filled for each year of CapEx at today’s levels.”

Cahn recently updated his math to see if this $200 billion hole still exists. He found that it tripled. 

Source: Sequoia

Again, what we have here is a bubble — the cost of AI (in chips, datacenters, energy, etc.) is enormous. And while revenue is beginning to grow, it is very far away from a point just of break-even. 

The issue is in value creation. Considering that Large Language Models are known to be unreliable at best (while potentially jeopardizing data security at the same time), the value of using them is, at the very least, shaky. Pointing out the significant value consumers get from their $15/month Netflix subscription, Cahn said that “long term, AI companies will need to deliver significant value for consumers to continue opening their wallets.”

  • He added that the winners in this arena will be the companies that focus on delivering actual real-world value to end users. 

  • I’ve spoken with numerous VCs who have similarly said that this question of real value (transformative enough to get people to pay for it) is what will decide which companies will survive the bursting of this bubble. 

“Those who remain level-headed through this moment have the chance to build extremely important companies,” Cahn said. “But we need to make sure not to believe in the delusion that has now spread from Silicon Valley to the rest of the country, and indeed the world … that we’re all going to get rich quick, because AGI is coming tomorrow.”

“In reality, the road ahead is going to be a long one. It will have ups and downs. But almost certainly it will be worthwhile.”

Which image is real?

Login or Subscribe to participate in polls.

A poll before you go

Thanks for reading today’s edition of The Deep View!

We’ll see you in the next one.

Your view on digital necromancy:

Around a third of you said it’s fine as long as there is consent. Another 25% of you said it’s a good way to deal with loss, while 25% of you said it’s wrong.

Let them go:

  • “It's creepy! Let the dead rest. Whether it's Judy Garland or your parents, let go of them. Recreating their voice or even their likeness in some fake form is exploitative and another money-hungry game. Sad, just plain sad."

It’s okay as long as families/individuals consent:

  • “I could see using my late mother's voice for something..."

All other things being equal, would you switch from Google Drive/other services to alternatives that guarantee privacy, security and no AI training?

Login or Subscribe to participate in polls.