‘AI progress’ drives strong quarters for Meta, Microsoft

October 31, 2024

Good morning, and Happy Halloween.

Whether you’re in costume, prepping someone else’s costume, or answering the door to shower strangers with sugar, hope you all have a fun holiday.

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

In today’s newsletter:

🕷️ AI for Good: Lyme disease detection
🚘 Waymo’s steady growth
💻 OpenAI open-sources factuality benchmark
📊 ‘AI progress’ drives strong quarters for Meta, Microsoft

Source: UCLA

Cases of Lyme disease have been on a steady incline for more than 20 years. But scientists aren’t exactly sure why, and doctors tend to have a tough time both diagnosing and treating Lyme.

This is partly due to the fact that early-stage symptoms can be similar to a bunch of other diseases, and partly due to the fact that testing for Lyme isn’t a quick or simple process; it involves two stages of testing based on lab work, and can take up to two weeks.

What happened: Researchers at the University of California, Los Angeles (UCLA) recently designed a quick, at-home Lyme testing kit. It’s made possible by artificial intelligence.

Somewhat similar in concept to an at-home Covid-19 test, this Lyme detection kit requires a blood serum sample which is dropped in a cartridge loaded up with special paper and proteins. The blood flows along the paper, creating a pattern which is then read by an AI algorithm.
Trained on samples from the Lyme Disease Biobank, the researchers reported a 95.5% sensitivity in detecting Lyme and a 100% specificity in ruling out the disease.

The digital reader was adapted from a $200 smartphone.

Why it matters: For antibiotics to eradicate Lyme, they have to be taken shortly after a tick bite. After a certain period of time, Lyme can become chronic, leading to a long list of health risks and issues. “A lot of folks find out they have Lyme disease well after the point at which they could have been treated very easily,” co-corresponding author Dino Di Carlo said.

The team is working on scaling up the tech and hopes it can be deployed in clinics in a few years

We live in an online world with little privacy and tons of threats. Surfshark is the best way to ensure that you're protected.

Whether you're looking to browse ad-free, protect yourself from viruses or surf the web without being tracked, Surfshark has you covered.

Surfshark performs real-time credit and ID alerts, plus they provide instant email data breach notifications and regular personal data security reports.

Protect yourself online with Surfshark. Get 80% off when you sign up today.

Source: Waymo

Self-driving unit Waymo reported Tuesday night that it is now delivering 150,000 paid trips each week, for a total of one million weekly autonomous miles.

The details: “The future is here,” Waymo said, “it’s growing, and it’s taking riders safely to their destinations every day.”

The expansion comes just two months after Waymo announced that it was delivering 100,000 paid trips each week, then a massive milestone.
It also comes very shortly after Waymo closed a massive $5.6 billion funding round — led by parent company Google — which it will use to continue expanding its operations.

As of September, Waymo had a total fleet of 700 driverless cars, each equipped with a stack of sensors and technology; a single Waymo has been reported to cost around $200,000.

While Waymo has shared details on 22 million driverless miles — which it claims were safer than humans over that same distance — it is not at all clear if that safety will scale up as Waymo does so.

What is clear is that Waymo is far ahead of the competition. By the time Tesla gets a robotaxi out there — if it ever manages to do so — Waymo may well possess entrenched dominance in the sector.

The AI consulting market is about to grow by a factor of 8X – from $6.9 billion today to $54.7 billion in 2032. But how does an AI enthusiast become an AI consultant? How well you answer that question makes the difference between just “having AI ideas” and being handsomely compensated for your contribution to an organization’s AI transformation.

Thankfully, you don’t have to go it alone – our friends at Innovating with AI have welcomed 300 new students into The AI Consultancy Project, their new program that trains you to build a business as an AI consultant. Some of the highlights current students are excited about:

The tools and frameworks to find clients and deliver top-notch services
A 6-month plan to build a 6-figure AI consulting business
Students getting their first AI client in as little as 3 days

And as a reader of The Deep View, you have a chance to get early access to the next enrollment cycle.

Click here to request early access to The AI Consultancy Project

AI enterprise startup Sierra raised $175 million at a $4.5 billion valuation.
GPU provider GMI Cloud raised $82 million in Series A funding.

How OpenAI CEO Sam Altman’s lobbying power tamed Washington (CNBC).
Waymo wants to use Google’s Gemini to train its robotaxis (The Verge).
How X users can earn thousands from US election misinformation and AI images (BBC).
Myanmar residents struggle to overcome severe internet blackouts with Starlink (Rest of World).
Flash floods kill at least 95 people in Spain (NBC News).

If you want to get in front of an audience of 200,000+ developers, business leaders and tech enthusiasts, get in touch with us here.

Director of Product Management: Duolingo, New York, NY
Head of AI Research: Leonardo AI, Remote

Deepgram: A platform for speech-to-text and text-to-speech conversion.
Depthify.ai: A tool to convert 2D videos into 3D spatial videos.

Source: OpenAI’s SimpleQA results

A massive problem with Large Language Models (LLMs) is their tendency to hallucinate, which refers to the generation of factually incorrect output. There remains no indication that LLM hallucination is solvable, something that ought to establish massive barriers between general deployment and legitimate use, especially in high-risk scenarios.

Still, there are a few different ways to mitigate hallucinations. One of these methods involves an attempt to learn how trustworthy — or factual — a given model is.

What happened: OpenAI on Wednesday released and open-sourced SimpleQA, a benchmark covering a wide range of topics designed to assess model factuality.

The benchmark is highly limited in scope, focusing exclusively on short queries with a single, verifiable answer. Still, OpenAI said that, under these constraints, it is accurate.
Across the given dataset, GPT-4o mini answered less than 9% of questions correctly. OpenAI o1 — OpenAI’s most powerful model — answered 42.7% of questions correctly.

The benchmark also allowed OpenAI to test model calibration; i.e., whether the model somehow knows when it is correct. This is achieved by requesting confidence scores for each answer. OpenAI found, however, that all of its models “consistently overstate their confidence.”

Why it matters: If verifiably accurate, confidence scores could introduce a whole layer of trust to these inherently untrustworthy systems. The problem, however, is multifold; the confidence scores aren’t there yet, and the systems aren’t there yet, either.

It’s important that even with o1, OpenAI’s “reasoning” model, the rate of factuality was below chance. Even in the best models, and even after years of focused study and billions of dollars, hallucination has not been solved; these models should not be used in critical situations.

Source: Meta

In round two of Big Tech earnings this week, heavyweights Microsoft and Meta reported third-quarter earnings Wednesday night. The major stock indices — the S&P 500, Dow Jones Industrial Average and Nasdaq — all closed the day down slightly in the lead-up to the reports.

Artificial intelligence was a main focus for both companies headed into the report.

Here’s how they did.

Microsoft:

Microsoft reported revenue of $65.59 billion, a 16% year-over-year increase that came in above the $64.51 billion that the analysts were expecting.
The tech giant reported earnings of $3.30 per share, well above the $3.10 number that analysts were anticipating. Net income for the quarter was $24.7 billion, an 11% increase.

But the number everyone was anxious about was Azure growth. Microsoft’s Intelligent Cloud revenue was $24.1 billion, a 20% increase from last year that was boosted by a 33% revenue growth in Azure and other cloud services. This is the most detailed clarification we have thus far received regarding Azure’s performance, an important indication of Microsoft’s AI success (or lack, thereof).

“AI-driven transformation is changing work, work artifacts and workflow across every role, function and business process," CEO Satya Nadella said. “We are expanding our opportunity and winning new customers as we help them apply our AI platforms and tools to drive new growth and operating leverage.”

During a conference call with analysts, Nadella said that demand for Microsoft’s cloud infrastructure is higher than Microsoft’s current capacity.

The stock, up about 15% for the year, fell 3.7% in after-hours trading as of Wednesday night.

Deepwater’s Gene Munster said it “was all about Azure and they delivered.”

Meta:

Meta reported total revenue of $40.59 billion, a 19% year-over-year increase that came in very slightly above analyst expectations of $40.29 billion.
The social media giant reported earnings of $6.03 per share, above analyst expectations of $5.25. Guiding fourth-quarter revenue somewhere between $45 billion and $48 billion, Meta said its full-year capital expenditures will be between $38 billion and $40 billion, a slight increase from its prior range.

"We had a good quarter driven by AI progress across our apps and business," CEO Mark Zuckerberg said. During Meta’s earnings call with analysts, he warned that Meta’s “AI investments continue to require serious infrastructure, and I expect to continue investing significantly there, too.”

Meta said that 3.29 billion people use a Meta service every day.

The stock, up 67% for the year, fell around 3% in after-hours trading.

Munster said that, for Meta, the key takeaway is “the core business is doing great and they’re aggressively investing in AI to become the next big thing. My guess is that big thing is related to monetizing search.”

Every tech earnings season recently has been about the AI trade. The bigger picture questions are focused on the ecosystem; is demand strong? Is demand getting stronger? Is it weakening?

As long as these tech giants keep boosting their capital expenditures to build out more infrastructure to handle the potential demand for AI, the “AI trade” will continue, with the Magnificent Seven pouring billions into Nvidia for the promise of an eventual return on their investment.

But the days of a tech giant saying “AI” during an earnings call and experiencing an enormous stock boost are over. Wall Street wants evidence that it’s truly paying off. That evidence just isn’t here yet.

Nvidia closed regular trading down around 1%, and fell another 1% in after-hours trading.