AI Daily Briefing — 2026-07-13

The push for open source artificial intelligence has never been more urgent, according to Hugging Face CEO Clem Delangue. In a recent interview on the Equity podcast, Delangue explained that his platform—often described as a GitHub for AI—has become essential for companies building and sharing open models. He noted that roughly half of the Fortune 500 now use Hugging Face, and many firms initially rely on proprietary APIs before switching to open source models to cut costs as they scale.

Delangue raised concerns about the growing concentration of AI power among a few major corporations, especially in light of Anthropic’s halted Fable release. He argued that open source development is vital to prevent a handful of companies from controlling the future of the technology. He also highlighted a geopolitical dimension: Chinese labs now produce most of the open models downloaded in the U.S., a trend he believes should be addressed by strengthening domestic open source efforts rather than abandoning the approach.

Beyond chatbots and coding tools, Delangue sees robotics as a field where open, transparent AI is especially critical. He warned that household robots will have intimate access to people’s homes and families, making it essential that the underlying models are open to public scrutiny. Hugging Face itself has prioritized capital efficiency over aggressive fundraising, notably turning down a large investment from Nvidia last year.

For the full discussion, including how Hugging Face is navigating these challenges, listeners can tune into the Equity podcast episode featuring Delangue.

Open source artificial intelligence is experiencing a surge in adoption as businesses move away from costly proprietary APIs, according to Hugging Face CEO Clem Delangue. His platform, often described as a GitHub for AI development, now serves roughly half of the Fortune 500, offering a space where builders share and download open models and datasets. Delangue notes a recurring pattern: companies initially rely on frontier APIs but quickly pivot to open source as scaling drives up expenses.

In a recent interview on TechCrunch’s Equity podcast, Delangue discussed the broader implications of the open versus closed source debate, particularly after Anthropic paused its Fable release. He expressed concern that a small number of major corporations could eventually dominate the AI landscape, stifling innovation and access. The shift toward open models, he argued, is not just a cost-saving measure but a strategic move to maintain control and flexibility.

Delangue’s observations highlight a growing tension in the industry. As more enterprises seek autonomy over their AI tools, the demand for open source solutions continues to climb. This trend, he believes, could reshape the competitive dynamics of the sector, challenging the dominance of proprietary systems and fostering a more decentralized future for artificial intelligence.

Anthropic has achieved the clearest view yet of what happens inside a large language model as it processes questions. Using a new tool called the Jacobian lens, researchers uncovered a hidden space—dubbed J-space—within its flagship model, Claude. This area contains words related to responses the model considers but may not ultimately produce. If Claude were human, these hidden words would reveal what it's thinking before it speaks, offering both fascinating and unsettling insights into AI reasoning.

Meanwhile, OpenAI has launched its long-awaited "super app" for ChatGPT, blending its chatbot, coding tools, and new models into a single workspace. Designed to work alongside users, the platform arrived on the same day as OpenAI's GPT 5.6 models. The company is also developing a fully automated researcher, signaling a push toward more autonomous AI assistants.

In other developments, humanoids performed the first teleoperated surgery on living animals, removing gallbladders from pigs. SK Hynix secured the largest US listing by a foreign company, raising $26.5 billion amid surging AI demand. Tencent is leading a deal to unwind Meta's $2 billion acquisition of Chinese AI startup Manus, while researchers revived human retinas hours after death, advancing eye transplant possibilities.

Source: MIT Technology Review AI

Anthropic has developed a technique offering an unprecedented view into the inner workings of large language models. By creating a tool called the Jacobian lens (J-lens), researchers uncovered a hidden space—dubbed J-space—inside Claude Opus 4.6. This space contains words related to responses the model may produce in the near future, revealing what it is effectively “thinking” before it speaks. The company says monitoring these words provides a new way to understand and control AI behavior.

The J-lens builds on prior mechanistic interpretability research, which MIT Technology Review AI named a breakthrough technology this year. It adapts an older tool called the logit lens, which identifies words an LLM is likely to output next. The J-lens goes deeper, picking out words the model may use later in its response. This reveals that LLMs often compute many possibilities beyond the immediate next word, some of which never appear in the final output.

Anthropic shared its findings in a paper and partnered with Neuronpedia to create an interactive demo. Tom McGrath, chief scientist at Goodfire, called the work “very good and interesting.” The technique marks a significant step in mechanistic interpretability, offering clearer insight into the mysterious middle layers where LLMs perform their heaviest computation.

Four U.S. nuclear reactors achieved criticality just in time for the nation's 250th birthday, surpassing a goal set by the Trump administration to have three microreactors reach this milestone by July 4. Criticality means a reactor can sustain a chain reaction, marking a technical achievement for nuclear power amid growing demand for clean electricity. However, reaching this stage does not mean the reactors are ready to supply power to the grid, as further development is needed before they can generate electricity.

In other tech news, China is reportedly planning to allow its top AI firms, including Alibaba, ByteDance, and DeepSeek, to purchase Nvidia H200 chips. This move comes after the U.S. authorized such sales, though China had previously withheld approval. The chips are critical for advancing AI capabilities, and this decision could reshape competition in the sector.

Meanwhile, NATO is building a network of sensors, drones, satellites, and AI to detect and stop Russian cyberattacks. Researchers are also exploring solar geoengineering to mitigate future El Niños by deflecting sunlight, though concerns about unintended consequences persist. Meta is patenting an AI device that records users to analyze emotions for personalized workout plans, raising privacy questions.

Chipmakers are stacking transistors vertically to maintain progress as Moore's Law slows, while a study shows Ivy League students suspected of AI cheating saw their in-person scores drop sharply. Other developments include detector satellites for nuclear bombs in space and a World Cup match driving Google Search traffic to a new record.