AI Daily Briefing — 2026-07-07

A new commercial from Google, released on the 250th anniversary of the Declaration of Independence, humorously asks what would happen if the Founding Fathers had access to modern workplace technology. The ad, titled with the tagline “Group project, but make it 1776,” shows Thomas Jefferson receiving a text from Ben Franklin while drafting the document. The collaboration unfolds through Google Docs, Calendar, and Meet, with attendees turning off their cameras, and ends with e-signatures and fireworks.

The fictional founders use Google’s AI tools throughout the process. They employ the “help me visualize” feature to test different animals for the national seal, Gemini takes meeting notes, and the chatbot advises them on declining King George III’s document access request. Unlike previous controversial Google ads, this one avoids suggesting AI should write the actual Declaration text, instead keeping the AI integration relatively subtle and tongue-in-cheek.

Some viewers have noted that the footage itself appears to have an AI-generated quality, giving it an uncanny glow. While reactions on YouTube and Instagram have been mostly positive, critics on Bluesky have called the ad “cringey” and “tone deaf.” Historian Angus Johnston pointed out that “even in a corny fantasy joke, it’s impossible to make the case that AI is a useful tool for political organizing, writing, or human collaboration,” highlighting the ongoing debate about AI’s role in creative and historical contexts.

Mistral AI has found itself in the spotlight following recent geopolitical shifts and a growing demand for sovereign technology. The French company, often compared to OpenAI, is frequently misunderstood. While its chat platform, Vibe, lacks the brand recognition of ChatGPT or Claude, Mistral is not trying to be Europe's answer to those giants. Instead, it is following a model similar to Palantir, deploying engineers to help governments and large corporations adopt and customize AI for their specific needs.

This enterprise-focused strategy suits Mistral's scale. Although the company is reportedly raising $3.5 billion at a $23.15 billion valuation—far less than U.S. frontier labs—its revenue has skyrocketed. Annual recurring revenue hit $400 million in February, up from $20 million the year before, and the company expects to surpass $1 billion this year. This growth has earned Mistral a seat at global tables like Davos and the French Parliament.

CEO Arthur Mensch recently outlined the company's vision in a LinkedIn post, emphasizing that Mistral exists to provide access to the best AI systems outside centralized control. The company is investing heavily in research, promising an open-weight model this summer. It also acquired infrastructure startup Koyeb and announced a €4 billion investment to build data centers in France and Sweden, reinforcing its commitment to sovereign AI.

Despite the buzz around its upcoming model, Mistral's most significant moves may be behind the scenes. By focusing on enterprise deployment and sovereign infrastructure, the company is carving out a unique position in the AI landscape—one that doesn't require beating OpenAI at its own game.

In South Korea, semiconductor employees are suddenly the most sought-after partners in the dating scene, thanks to massive bonuses tied to the AI chip boom. Baek, a 35-year-old manager at SK Hynix, says his mother signed him up with a matchmaking agency a year ago, a common move among anxious Korean parents. Recently, however, Baek and his colleagues have noticed a surge in interest from potential dates, likely due to the eye-popping profits their company is sharing. SK Hynix and Samsung have both struck deals to pay workers a portion of operating profits, with some employees receiving nearly $500,000 extra this year alone.

The AI chip boom has transformed South Korea’s social landscape, creating a new elite of “silicon-collar” workers who earn roughly 20 times the national average. These employees are spending freely on luxury goods, homes, and matchmaking services. Lee Sung-mi, a matchmaker at Seoul-based Sunoo, says clients who once rejected chip workers are now asking to be reintroduced. One woman who previously turned down an SK Hynix employee because his factory was too far from Seoul recently asked for a second chance; they’ve been dating for a month.

While the windfall is helping chip workers find love, it is also fueling public debate about deepening wealth inequality. South Korea’s economy increasingly revolves around Samsung and SK Hynix, which supply most of the world’s high-bandwidth memory chips for AI systems. As demand for these chips skyrockets, the gap between semiconductor employees and other workers grows, raising concerns about social division in a country already grappling with economic disparities.

Most large language models suffer from a groupthink issue, producing overly predictable responses. For instance, asking a chatbot for a random number between 1 and 10 almost always yields 7. While this consistency benefits coding or research tasks, it stifles creativity in areas like brainstorming or travel planning. Australian startup Springboards aims to break this pattern with its new LLM, Flint, designed to generate more diverse answers to open-ended questions, such as suggesting varied European destinations. The company is pushing chatbots beyond the obvious, according to MIT Technology Review AI.

Flint's training focuses on expanding the range of possible outputs, offering users richer and less repetitive suggestions. This approach addresses a growing frustration with mainstream AI models that tend to default to common or safe responses. By prioritizing variety, Springboards hopes to make AI more useful for creative and exploratory purposes, where novelty is key.

The development comes as the AI industry grapples with balancing reliability and innovation. While predictable models serve structured tasks well, they fall short in scenarios requiring out-of-the-box thinking. Springboards' solution represents a shift toward more dynamic AI interactions, potentially influencing how future models are designed.

As the field evolves, overcoming groupthink could unlock new applications for AI in entertainment, education, and personal planning. Springboards' Flint is an early step in that direction, challenging the status quo of chatbot behavior.

A recent, quietly implemented update to Google’s privacy settings means that any media you upload—such as images, audio recordings, or video clips—may now be used to train the company’s artificial intelligence models. The change, announced via email in June, automatically opted users into this expanded data collection under the premise of offering more control over personalized recommendations and saved history. This applies not only to Google Search but also to services like Maps, Shopping, Flights, Hotels, Translate, and News. For instance, photos taken with Google Lens or voice recordings from Search Live or Translate can now be retained for AI development.

Google confirmed in its email that saved media is used to “develop and improve Google services and technologies, including AI models and safety measures.” The company’s help documentation adds that this data may also be reviewed by human reviewers. While some storage is temporary for product functionality, Google explicitly states that saved media can be kept specifically for AI training purposes. This shift reflects a broader industry trend, with companies like Meta also collecting user-uploaded content to fuel their AI systems.

Fortunately, users can opt out. Navigate to the Search Services History page and uncheck the “Save Media” box separately from the “Search Services History” box. You can also set automatic deletion intervals of 3, 18, or 36 months. For further control, visit the Web & App Activity settings to manage other data like location and browsing history. Taking these steps ensures your personal media is no longer used to train Google’s AI without your explicit consent.