🔮 Can Google transform?; driving the energy transition; antagonistic AI, GRU spies and robots++ #463

Hi, I’m Azeem Azhar. In this week’s edition, we explore

• The prospects of Google and Apple to transform themselves. The two firms have different track records and the task may be harder for one.

• The implications of the early results of the effectiveness of LLM-based customer support.

• The superheroes accelerating the net-zero transition… and more.

---

Sunday chart: A tale of two titans

The recent Gemini hiccup ignited a politically polarised firestorm, yet the incident highlights an even greater problem for Google: its sensitivity to missteps. A culture of conservatism could be seeping into a company that burst into our lives on the back of breakthrough science, per Larry Page and Sergey Brin’s work at Stanford. One analysis compared Google’s current state to IBM’s historical trajectory. Once a pioneer, IBM eventually became a large, established entity struggling to innovate and forge new paths.

Google’s revenues are intrinsically tied to advertising, its core business model. In some sense, it is a media company. Might the cash cow have made Google sluggish?

For the first time in years, the tasks we’ve delegated to search engines can go somewhere else and often better – answer engines like Perplexity or You.com, for example. My Perplexity and ChatGPT usage is roughly half the level of my Google usage (I discussed this with Perplexity co-founder Aravind Srinivas).

Analysts at Gartner reckon on a 25% drop in search engine queries by 2026 due to generative AI, representing a massive potential revenue loss for Google. Not just that, its advertising business has huge margins. Google’s overall net margin is 34% and the ad business is carrying all of Google’s loss-making bets. 

Google’s challenge now is to see if its culture can do what it hasn’t done in the past two decades: find a significant, profitable revenue line to diversify its business and sprout new sources of growth. 

That distinguishes it from Apple. The firm shut down Titan, its electric vehicle project. This was no small call – with 2,000 people and around $10 billion invested so far. Apple’s judgement probably says something about the prospects for level five autonomy – not great.  

But Apple has proved it can make successful adjacent bets, shifting its revenue mix from computers to phones and services. 

The iPhone meant Apple needed to skill up on a new form factor, hardware device and set of relationships (with carriers). Yet it is still built on Apple’s core competencies in software and customer experience. And more recently, Apple has transitioned into services: tied, ultimately, back to the user base of its iOS platform. Though services only represented 22% of Apple’s 2023 revenue, they contributed an even higher 35% towards profits, driven by their strong margins. For example, Apple placed a shrewd adjacent bet with Apple Pay, essentially swapping the phone’s traditional pocket companion – your wallet – for itself. 

And that might help us understand other rationales for the cancellation of the car project. Cars are far from an adjacent product category for Apple. Few of its core capabilities would translate to vehicles. And cars are a low-margin business, even at the luxury end (BMW’s net margin is 7%.) The car industry’s shape has changed since the Titan kicked off: Tesla’s lost momentum, Chinese manufacturers are driving prices down and traditional players are getting better at delivering EVs. Integration for iOS will, for most of them, still be desirable. The markets agree: they shrugged off Titan following the fate of the Titanic, probably relieved to see those resources allocated to a technology that could make Apple money: AI. 

Google struggles where Apple succeeds: doggedly pursuing transformative adjacent bets. Google makes technical breakthroughs but has yet to bring them to market. It invented the transformer architecture, only to watch OpenAI capitalise on it. Now it frantically plays catch-up. Even Google Cloud, its most successful bet, feels disparate – a world apart from the consumer-facing marketing that defines Google’s multicoloured media brand. This lack of adjacency could hinder truly transformative potential.

Google is fast approaching its crossroads. Which way will it go?

---

Key reads

Klarna Klarna Klarna Klarna’s [Customer Support] Chameleon. Klarna’s AI chatbot is making waves. Klarna announced in a press release that in just one month, its AI chatbots handled 66% of customer enquiries (over 2.3 million chats). Apparently working five times faster than humans – and displacing 700 workers. These bots are currently functioning as rather simplistic but highly successful filters. They’re likely to rapidly evolve to be able to tackle more complex queries (see, for example, the startup Sierra). As their capabilities improve, companies will invest further, streamlining operations and potentially reducing reliance on traditional customer support structures. While companies see the potential for efficiency gains, a safety architecture for displaced workers involving economic buffers and skills support could become crucial. About three million people work in call centres in the US. 

The new eco-warriors. Back in September last year, Michael Liebreich introduced us to the “Five Horsemen” that challenge the achievement of net-zero emissions. After keeping us in suspense for five months, he revealed “the five Superheroes even more powerful than the Five Horsemen”. This is an excellent framework for understanding the obstacles and opportunities within the energy transition. His first superhero is a familiar one to readers: Exponential Growth. His second superhero highlights another powerful yet overseen interplay between energy and information. We now possess the capacity to dramatically increase information flow within our energy systems by integrating renewable energy sources with intelligent demand response programs and grid management solutions. This approach makes tackling the challenges of intermittency associated with renewables significantly more feasible.

Au large concern. Mistral AI released its new AI model “Au Large” this week, sparking criticism for two reasons. Firstly, despite being known for open-source advocacy, Mistral chose to make the model closed-source. Secondly, its new cloud partnership with Microsoft raises concerns about Mistral’s independence and potential alignment with big-tech interests during discussion of the European AI Act. This move highlights the growing influence of the “Cloud Keiretsu” on public goods (as highlighted in EV#442), a worrying trend for some in Europe. Mistral, once seen as Europe’s answer to US dominance in AI, now appears to be increasingly influenced by US capital. This reflects Europe’s broader challenge in competing technologically. As Yann Coatanlem notes, while European regulations excel at promoting welfare, their structural rigidity may hinder companies’ abilities to adapt quickly to new technological realities.

---

Newsreel beta

This new section we’re testing contains important news items about AI and exponential technologies for the week.

• Apple’s eighth- and tenth-largest shareholders, Norway’s wealth fund and LGIM, will support a resolution requesting an AI ethics report at Apple’s annual meeting.

• Elon Musk sues OpenAI over its decision to move to making closed-source models.

• Figure, a humanoid robotics company, raises $675 million from investors including OpenAI, Nvidia and Microsoft.

• Meta is planning to launch Llama 3 in July.

• Germany aims to go beyond achieving net-zero emissions by 2060 by setting a target for net-negative emissions.

• Genie, an AI by Google, can build playable virtual worlds from an image.

• E-commerce giant Alibaba leads a $1 billion investment in Moonshot AI, valuing the Chinese genAI startup at $2.5 billion.

• Intel is planning to manufacture 1nm chips in 2027.

---

Data

By incorporating borrowing costs, researchers found they could explain up to three-quarters of the discrepancy between current consumer sentiment and what traditional economic indicators, like unemployment and inflation, suggest it should be.

The video game industry saw record-breaking reported layoffs in both 2022 (8,500) and 2023 (10,500), with 2024 already surpassing 8,000. This comes after the industry’s revenues contracted last year.

A study highlights that LLMs with Retrieval Augmented Generation (RAG) struggle to find and verify medical sources, often generating irrelevant ones. The best, GPT-4 with RAG, lacked source support for 30% of its claims.

Microsoft shows that 1-bit LLMs are more efficient than traditional systems. They can use up to 98% less energy, respond up to 310% faster and use as much as 85% less memory while performing at the same level of quality. (Simple video explainer here.)

Between 2019 and 2023, clean energy has grown nearly twice as fast as fossil fuels. 

The private sector invested $5-6 for every $1 invested in clean energy and transportation by the US federal government in 2023. 

UBS says that to justify Apple’s current valuation, everyone needs to spend $800 a year on Apple products.

---

Short morsels to appear smart at dinner parties

📝 The quality of personality tests varies greatly: the usefulness of the MBTI is halfway between science and astrology.

😧 Hair raising: Jan Marsalek, COO of Wirecard, which collapsed in scandal in 2020, is believed to have been a GRU spy. 

📍 The surprising story of how targeted ads became one of the most precious sources of geospatial intelligence.

🤖 The incredible one-year improvement in the walking speed of the Tesla Optimus robot.

😡 Researchers are exploring how antagonistic, confrontational and rude AI could be useful to humans. 

😶‍🌫️ Riders in the smog: Pollution monitors worn by gig workers expose extreme air pollution levels in South Asia.

---

End note

I’ve got a new website outlining my speaking work. I’ve given keynotes at more than 100 organisations on AI, the exponential transition and, well, all the things you’ve enjoyed in this wondermissive. Check it out.

❤️ We are changing a couple of things in the newsletter to improve your experience. Let us know how we’re doing:

What did you think of today’s edition?

Excellent | Very Good | Good | Fair | Poor

Cheers!

A

---

What you’re up to – community updates

• Kevin Delaney’s company, Charter, which specialises in future-of-work reporting and research, has secured funding from FT Ventures.

• Francesca Warner participated in a government task force aimed at finding ways to accelerate the growth of women-led businesses. The task force has now released its final report.

• Sarah Gold, together with her co-authors, has published six practical and pragmatic recommendations for Britain’s next government regarding public service delivery and technology.

• Simone Vannuccini co-authored a research paper that examines how the European Union can develop a supranational industrial policy amid the age of polycrisis and permacrisis, particularly within the framework of escalating international rivalries over critical technologies.

• Dan Stanley has written on how implicit cultural narratives about technology and data are influencing our reaction to AI.

• David DeLallo and co-authors have written an article for Harvard Business Review on integrating genAI into your organisation.

• Euro Beinat and his team created a helpful A-Z guide to the terminology of artificial intelligence.

Share your updates with EV readers by telling us what you’re up to here.