(#74) Apple's AI strategy; Google O/I 2024; Starbucks drinks need less steps to make; Open AI has a new model

Marc Andreessen on why building companies will become more expensive in the AI Era

May 16, 2024

Shape the Future: Share Your Insights in Our Survey!

Share your insights on the importance of upskilling and reskilling in the AI era. Our survey aims to better understand the dynamics and barriers of today's agile workplace.

Your input is essential for our study and the survey takes only 7 minutes to complete.

Click here to start the survey.

Thank you for your time!

Strategy

Tariffs for Chinese EVs and more
President Joe Biden is increasing tariffs on Chinese-made electric vehicles, solar panels, steel, and other goods. The White House announced a 100% border tax on electric cars from China as part of measures to counteract what it considers unfair trade practices and to protect U.S. jobs. China opposed these hikes and promised retaliation. Analysts suggest the tariffs are largely symbolic, aimed at bolstering support in a challenging election year. This follows criticisms from former President Donald Trump, who argued Biden's policies would harm the U.S. auto industry. Biden asserted the need to prevent China from dominating the market for electric vehicles and other essential goods. The new tariffs, affecting $18 billion worth of imports, include significant increases in electric vehicle tariffs and levies on solar cells and certain metals. Despite China's warning of negative impacts on bilateral cooperation, the Biden administration emphasized that these measures are crucial for ensuring a secure supply of critical products domestically, reflecting a shift in U.S. trade policy towards protectionism.

New tariffs - at a glance

semiconductors - from 25% to 50% by 2025
certain steel and aluminum products - from 7.5% to 25% in 2024
electric vehicles - from 25% to 100% in 2024
lithium batteries and critical minerals - from 7.5% to 25% in 2024
solar cells - from 25% to 50% in 2024
ship to shore cranes - from 0% to 25% in 2024
rubber medical and surgical gloves - from 7.5% to 25% in 2026

Other insights:

1/ Japan, South Korea, and Germany were friend countries, while China is a competitor. With business partners, you can negotiate, with competitors not that much

2/ Japan, South Korea, and Germany have opened factories across the USA, something that will not happen with the Chinese automakers, although they will open (and force) market entry through Hungary (in Europe) or Mexico (in North America)

3/ Reliance on Chinese software (including 5G) is a national threat, hence the act.

4/ What is more important is that The White House referred to the “Made in China 2025” policy several times. “Made in China 2025” should be read by any business students or business professionals to better understand what’s a problem at the core. Tariffs, The White House, Made in China 2025

Starbucks shares sink 12% amid poor sales
Going through the articles and CEO post and replies I understand that one of the main core problems is long waiting time.

Solution: Simplify the steps in making the products (including developing in-house machinery to do some of those steps). People are not coming to Starbucks for craft coffee.

But, can you do that when you have to prepare Pumpkin Spice Latte or Nitro Cold Brew Coffee? Sales, Rules, Blog

Google I/O 2024
What was it about? AI, of course. It was pronounced 121 times in less than 2 hours.

Here is what they announced:

1/ Google Lens now lets you search by recording a video

2/ Gemini will answer questions about your photo library

3/ Google’s flagship Gemini model gets faster and more capable
4/ Gemini joins users in Workspace

5/ Project Astra is Google’s star in AI

6/ Veo is a video generation based on text

7/ Gemini becomes a better conversation partner

8/ Google Chrome is getting an AI assistant

…and many more. LINK

Google is just now serious about AI and how it can use it. Probably, they will be a $3 trillion company by the end of the year.

OpenAI releases GPT-4o, a big step towards a frictionless world
OpenAI has introduced GPT-4o, an enhanced version of the GPT-4 model that powers ChatGPT. This new model is significantly faster and improves text, vision, and audio capabilities, according to OpenAI CTO Mira Murati. It will be available for free to all users, with paid users benefiting from increased capacity limits. The rollout of GPT-4o’s capabilities will be gradual, starting with text and image features. OpenAI CEO Sam Altman highlighted that GPT-4o is natively multimodal, meaning it can handle commands and generate content in various formats such as voice, text, or images. Developers can access the API, which is now offered at half the price and twice the speed of GPT-4 Turbo. New features for ChatGPT’s voice mode are also being introduced, enabling real-time interaction and environmental awareness, similar to a Her-like voice assistant.

Sam Altman, in a blog post, reflected on OpenAI’s evolution, acknowledging a shift from the original vision of creating widespread benefits to focusing on making advanced AI models available to developers through paid APIs. This shift has faced criticism for not open-sourcing the models, but Altman emphasized that the goal is for third parties to use these models to create innovations that benefit everyone. Before the GPT-4o announcement, there was speculation about OpenAI launching an AI search engine or a new voice assistant. The timing of this launch, just before Google I/O, indicates a strategic move by OpenAI as Google is expected to unveil various AI products at the conference.

Some takeaways:

1/ The speech synthesis is impressive and reminiscent of Google Duplex, which never gained widespread adoption.

2/ However, if OpenAI had developed GPT-5, they would have showcased it by now.

Despite 14 months of effort, GPT-5 has not been released.

3/ The most critical figure in the blog post is shown below, indicating that 4.0 is not significantly different from Turbo, and Turbo is not vastly different from 4.

4/ It seems OpenAI has shifted focus to new features because they are struggling to achieve the capability advancements that "exponential improvement" would suggest.

Crucially, each day without a GPT-5 level model from OpenAI or their well-funded, motivated competitors suggests we may be entering a phase of diminishing returns. GPT 4o, The Verge, Demo, Sam Altman

Apple’s new iPads
Ipads today make Apple around $10 billion per quarter, approximately as the Macs. No, people are frustrated that this piece of engineering is not used to the fullest due to its weak operating system. Somebody made a list of what’s missing on the iPadOS, but I think the list is useless because Apple said explicitly that will never put macOS on it.

The question I face is not about transforming an iPad into a Mac. Rather, it is about determining the ideal set of reimagined features that provide a modern productivity experience, with the assumption that compatible apps will follow. LINK

Industry loans in China
China's recent surge in industrial loans has been frequently interpreted as a sign of a renewed push in manufacturing. However, research from the Rhodium Group reveals that a substantial portion of these funds has not been directed towards manufacturing. Instead, the increased lending has primarily supported local government-related entities and financial speculation, leading to inefficiencies that could undermine China’s manufacturing growth. In the fourth quarter, loans to manufacturing companies accounted for only 63% of new industrial credit, a significant drop from 80% in early 2020. Analysts at Rhodium, including Endeavour Tian, highlight that China's domestic credit demand is weak, with much of the new credit being used for financial arbitrage and local debt restructuring rather than boosting manufacturing investment.

This trend poses a risk to China's strategy of using manufacturing to drive rapid economic growth post-real estate crisis, which may exacerbate trade tensions with the US and EU. Despite the rapid growth in medium- and long-term industrial loans since 2020, actual manufacturing investment has slowed. The Rhodium report indicates that many companies have diverted cheap bank loans into long-term deposits and investment products for profit. Consequently, the apparent rise in industrial loans masks a slowdown in manufacturing investment growth between 2022 and 2023. To address these inefficiencies, China's financial regulator has pledged to increase the proportion of mid- and long-term loans to the manufacturing sector and ensure that financial resources are more effectively utilized, as stated in recent government documents and comments from the People’s Bank of China. LINK

Artificial Intelligence

Apple AI Strategy
All eyes are on Apple at WWDC (June) to see what they will announce on AI. Here is my take:

1/ On-device Large Language Models (LLM) (In-house):

Apple is focusing on developing on-device large language models internally. By leveraging its in-house capabilities, Apple aims to enhance user privacy and data security since the processing occurs directly on the device without needing to transmit sensitive information to external servers. This approach also ensures faster response times and improved performance in real-time applications such as Siri, text prediction, and other personalized AI functionalities. On-device LLMs can be particularly useful for tasks that require immediate, offline processing, providing users with seamless and reliable AI-driven experiences.

2/ Cloud-powered Large Language Models (LLM) (In-house):

In addition to on-device LLMs, Apple is investing in cloud-powered large language models developed internally. These cloud-based models can handle more complex and resource-intensive tasks that exceed the capabilities of on-device processing. By utilizing its robust cloud infrastructure, Apple can offer powerful AI services that are scalable and accessible from multiple devices. This hybrid approach allows Apple to balance the benefits of on-device processing with the extensive computational power of the cloud, enabling sophisticated applications such as advanced natural language processing, large-scale data analysis, and machine learning tasks.

3/ Chatbot (OpenAI for sure, Google maybe):

Apple's strategy also includes integrating chatbots from external AI providers like OpenAI and potentially Google. By incorporating these advanced chatbot technologies, Apple can enhance its AI offerings with the latest innovations in conversational AI. OpenAI's models, known for their high performance and versatility, can significantly improve Apple's customer service, virtual assistants, and other interactive applications. While Google may also be a consideration, leveraging partnerships with established AI leaders allows Apple to deliver cutting-edge AI experiences without solely relying on in-house developments. This collaborative approach ensures that Apple remains at the forefront of AI advancements, providing users with state-of-the-art AI-driven interactions. OpenAI, Models, More LLMs

Marc Andreessen on why building companies will become more expensive in the AI Era
There is a common belief that AI will help reduce costs and enhance efficiency. However, Marc offered a compelling counterargument by citing Jevons' paradox.

Jevons' Paradox arises when technological advancements improve the efficiency of resource use, thereby reducing the amount needed for a particular application. However, the decreased cost of using the resource leads to such an increase in demand that overall resource use actually goes up, rather than down.

Marc provided several examples of Jevons' paradox:

Building more roads leads to increased car usage and traffic congestion.
CGI was developed to lower filmmaking costs, but higher audience expectations drove up CGI expenses.
During the Industrial Revolution, coal consumption rose as coal prices fell.

According to Marc, this paradox implies that while the cost of developing a particular piece of software may decrease, the significant rise in demand for more advanced software will ultimately raise the overall cost of establishing a software company. Customers will continuously seek more powerful features, driving up development costs. LINK

Things Happen

Singapore’s Prime Minister Lee Hsien Loong explains why he will never allow Western wokeism to take root in Singapore. LINK

The rise of the Manhattan mega renter, paying $25,000-$75,000 a month. LINK

America’s yogurt love affair is over. LINK

Are African prices too high? LINK

Podcast: Brad Setser, CFR senior fellow, in an interview on China's new tariffs with Jordan Scheiner. LINK

Foreign Policy has an in-depth analysis of India’s new idea under Modi. LINK

Beijing Auto Show 2024 was scary for Western journalists. A revolution is coming. LINK

Data

For many local air pollutants, we may have already passed the peak. LINK

The German approach to fight climate change:

Shut down nuclear power plants.
Continue operating coal mines and coal power plants.
Significantly underreport coal mine emissions, with a discrepancy factor of 184.
Render energy-intensive industries unprofitable in the process.

Electricity prices in Europe in 2023.

Outside Interest

New decade, new names in the USA. LINK

There is a large smuggling spare parts business to supply the Russians for aviation and other industries, mainly through the Emirates, Turkey, and Maldives. LINK

Google accidentally deleted a company's entire cloud environment and the backups. LINK

The world's fastest swordsman, Guinness World Record holder, Isao Machii. LINK

Interesting newsletters 💡

I will be a speaker at Techsylvania, held between 26-27 June in Cluj. Use the code SORIN24 to get a 24% discount on your tickets. LINK