Good morning from San Francisco!

While Elon Musk often takes a sledgehammer approach, OpenAI CEO Sam Altman takes a diplomatic and nuanced approach - recently upstaging Trump buddy Musk.

What happens when open-source AI is allowed to operate unchecked? The Chinese example of DeepSeek provides surprising, if shocking, insights: the so-called "wunderkind" not only generates instructions on how to make Molotov cocktails but also writes Nazi manifestos.

I wish you a good start into Monday,

Marcus Schuler

How OpenAI CEO Sam Altman outsmarted Elon Musk

Victory on points for Sam Altman. The New York Times smugly reports how the head of the San Francisco-based AI company stole the show from his rival Elon Musk on the day of Donald Trump's inauguration.

While tech leaders like Elon Musk and Mark Zuckerberg were in the spotlight at President Trump's inauguration, Sam Altman worked behind the scenes. Days before he flew to Washington, Altman had a 25-minute phone call with Trump in which he laid out a bold plan: In the Trump era, artificial intelligence would catch up to human intelligence. In addition, several specialized AI data centers would be built in the U.S. to leapfrog China in development.

Working with influential figures such as Oracle's Larry Ellison, SoftBank's Masayoshi Son, and North Dakota Governor Doug Burgum, Altman appears to have laid the groundwork for Stargate, which is being touted as the largest AI infrastructure project ever.

Why it matters:

• Although Altman had previously criticized Trump and clashed on several occasions with Musk, who supported Trump with more than $250 million in his campaign, Altman deftly positioned OpenAI at the center of the new administration's AI agenda.
• After the inauguration, Altman publicly credited Trump for the ambitious deal, eclipsing Musk's influence.
• This strategic maneuver secured significant investment for OpenAI. It redefined Silicon Valley’s power dynamic, proving that political acumen can be as valuable as technological innovation in the high-stakes race for AI.

Read on, my dear:

• New York Times: How Sam Altman Sidestepped Elon Musk to Win Over Donald Trump

AI-Photo des Tages

Prompt:
A hyper-realistic portrait of an old man with deep wrinkles, intense gaze, soft lighting, 85mm lens.

The dark side of DeepSeek: Chinese AI model open to dangerous jailbreaks

Unlike its Western competitors, the Chinese AI app DeepSeek produces dangerous content when jailbroken. AI security experts and the Wall Street Journal found that the R1 model generated instructions on modifying bird flu, a pro-Hitler manifesto, and a campaign for teenage self-mutilation.

Unlike U.S. developers with strict security measures, DeepSeek is vulnerable because of its open-source design. Researchers have been bypassing security measures to get tips on how to build weapons and malware and evade law enforcement.

Why it matters:

• Palo Alto Networks was able to generate step-by-step instructions on how to make a Molotov cocktail.
• CalypsoAI extracted instructions on how to evade law enforcement.
• Israeli cybersecurity firm Kela manipulated R1 to create malicious code.

Read on, my dear:

• Wall Street Journal: DeepSeek Offers Bioweapon, Self-Harm Information

Mind over keyboard: Meta's groundbreaking development of Brain-to-Text

Meta (formerly Facebook) has unveiled a groundbreaking lab experiment that turns brain signals into text. In a study of 35 volunteers at a Basque research center, a giant two-million-dollar magnetic resonance imaging (MRI) machine, similar to a lateral MRI, recorded neural activity as the participants "typed" in their heads.

A neural network then analyzed these signals and correctly identified up to 80% of the keys pressed by skilled typists. Although the device weighs half a ton, requires a shielded room, and is clearly unsuitable for everyday use, the research represents a significant advance in non-invasive brain reading.

More than a technological curiosity, the Meta project aims to decipher the brain’s hierarchical structure of speech production. The results of this study could help to develop future AI systems that learn and think more like humans, bridging the gap between neuroscience and artificial intelligence in an innovative way.

Read on, my dear:

• Meta research: Brain-to-Text Decoding: A Non-invasive Approach via Typing
• Meta research: From Thought to Action: How a Hierarchy of Neural Dynamics Supports Language Production
• Meta blog post: Using AI to decode language from the brain and advance our understanding of human communication

Better prompting with ChatGPT/DeepSeek

"How can I use DeepSeek to automate task? To save time, provide a step-by-step guide using prompts, Al tools, and workflows.”

Alexa 2.0 - Amazon's AI Relaunch in the Connected Home

Amazon's Alexa was once the undisputed poster child for intelligent home assistants. Since its launch in 2014, users have been able to use voice commands to set timers, play music, and get the weather forecast. By 2018, more than 100 million Alexa-enabled devices had been sold worldwide, and early users were thrilled with the benefits. At the height of its development, Alexa dominated the U.S. smart speaker market with nearly 70% share, but recent data shows that lead is shrinking.

However, with the rapid development of artificial intelligence, Alexa's skills were gradually overtaken. Competitors like Google Assistant and Apple's Siri quickly offered more sophisticated conversational skills and deeper software integration.

In the past two years, AI has exploded. OpenAI's ChatGPT, launched in November 2022, had more than 100 million users by early 2023, setting a new benchmark for natural language interaction. Today's AI models can write, program, and execute nuanced dialogues. On the other hand, Alexa seems antiquated and like a relic from the digital stone age. Meanwhile, Amazon has faced high-level layoffs and growing competition, forcing a strategic reassessment.

Now, Amazon is at a crossroads.

The company plans to overhaul Alexa significantly, integrating advanced language models and machine learning algorithms with its vast consumer data. A future Alexa should be able to provide accurate answers and engage in dialogue, and anticipate its users' needs. Such a transformation could secure Amazon at the forefront of the connected home technology market, which is expected to grow to 150 million devices worldwide by 2025, representing a potential revenue stream of $50 billion per year. A next-generation Alexa would redefine how we interact with technology and re-establish Amazon as a leader in consumer innovation.

Analysts predict the smart home market will exceed $50 billion in annual revenue by 2025, driven by consumer demand for more innovative, connected devices. The Alexa upgrade could be critical for Amazon to regain its leadership position in this rapidly evolving global market.

Read on, my dear:

• The Verge: Amazon’s new Alexa has reportedly slipped to 2025
• CNet: Alexa Is Set for Major AI Upgrade From Amazon. Here's What to Know
• CNBC: Amazon announces hardware event for Feb. 26 as it races to release new Alexa

AI & Tech News

Super Bowl ad as an AI experiment: When Veo 2 and ChatGPT Fail

Ari Kuschnir, founder of m ss ng p eces, is known for his viral AI experiments, from the Trump-Biden bromance to surreal political confrontations. Eager to capitalize on the Super Bowl's golden advertising moment, he investigated Google's new AI video generator, Veo 2. Veo 2 lacked audio support and character consistency despite its stunning visuals, limiting a continuous narrative.

Combined with ChatGPT's latest idea model, o3, Ari created a detailed shot list for a hypothetical Super Bowl spot meant to be funny and surreal. However, the end result seemed disjointed and incomplete, demonstrating that AI cannot yet produce compelling and original commercials. For now, AI remains a collaborative tool for human storytellers. The result was a surreal collage of short clips and a quirky song that showed the limits of AI.

Augmented Intelligence: Christie's AI exhibition divides the art world

Christie's, the auction house known for its art auctions, is holding its first exhibition exclusively devoted to AI-generated art in its Rockefeller Center galleries beginning February 20: "Augmented Intelligence".

The exhibition features works by artists such as Mat Dryhurst, founder of Spawning AI, and presents a robot that paints live. Critics, however, complain that many generative AI tools use artists' work without their permission and called for the project's cancellation in an open letter with over 3,000 signatures. Christie's defends the project, arguing that AI enhances each artist's existing practice by using data derived from their own input.

No more Silicon Valley: China's AI Stars Choose DeepSeek

DeepSeek attracts China's AI talent by recruiting graduates and interns from elite universities. Faced with visa problems and the high cost of living in the US, many young researchers are opting for jobs in their home country. Restrictions on chips in the US have further accelerated innovation in China.

After completing an internship at Nvidia in 2023, AI researcher Zizheng Pan quickly chose DeepSeek over a job in Silicon Valley. His decision reflects a broader trend among China's AI elite, who prefer competitive salaries, flexible working conditions, and an essential role in early career development at home.

DeepSeek's innovative culture and open access to IT resources are attracting the best minds to advance China's AI pioneering work.

The best tools: AI-based image generation

1. Midjourney: Creates images, paintings, digital art, logos, and more.
2. Replicate.com: Source models for machine learning using a cloud API.
3. V0.dev: UI generation using simple text commands and images